Football analytics: a literature analysis from 2010 to 2020
Leonardo Fontanive2021
visibility
…
description
51 pages
link
1 file
The overall goal for the current study is to present a literature review of analytics, precisely machine learning (ML) reference authors in terms of methods and applicable scopes of study, in football where is a field that historically there are empirical decisions and the usage of analytics has been growing intensely. The research aims to list relevant academic contributions published between 2010 and 2020, performing a comparable picture per authors across the following subsets: player individual technical skills and team performance. Furthermore, the approach will provide a summary of studies for machine learning methods applied in football. Such outcomes of this study would contribute to the discussion about football analytics. Regarding that these summaries can drive researchers to have a deep dive into the fields of interest straight to references preview studied in the thesis. Results indicate that football analytics has broadly vast opportunities in terms of research, regarding machine learning methods and a high potential to have a deep exploration of team and player perspective. This study can leverage and pavement new further in-depth and targeted investigation toward football analytics.
Related papers
Football(Soccer) Analytics: A Case Study on the Availability and Limitations of Data for Football Analytics Research
Football Analytics is fascinating area. This Case Study presents a unique situation where by a rating system for quantitatively measuring a player's performance was desired. This would eventually enable predictions derivations on various factors, such as player performance or match outcomes. Data for player rating had to be gathered from different sources; however, this Case Study outlines the solutions that were used to gather such data.
Machine learning application in soccer: a systematic review
Biology of Sport, 2023
Due to the chaotic nature of soccer, the predictive statistical models have become in a current challenge to decision-making based on scientific evidence. The aim of the present study was to systematically identify original studies that applied machine learning (ML) to soccer data, highlighting current possibilities in ML and future applications. A systematic review of PubMed, SPORTDiscus, and FECYT (Web of Sciences, CCC, DIIDW, KJD, MEDLINE, RSCI, and SCIELO) was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. From the 145 studies initially identified, 32 were fully reviewed, and their outcome measures were extracted and analyzed. In summary, all articles were clustered into three groups: injury (n = 7); performance (n = 21), which was classified in match/league outcomes forecasting, physical/physiological forecasting, and technical/tactical forecasting; and the last group was about talent forecasting (n = 5). The development of technology, and subsequently the large amount of data available, has become ML in an important strategy to help team staff members in decision-making predicting dose-response relationship reducing the chaotic nature of this team sport. However, since ML models depend upon the amount of dataset, further studies should analyze the amount of data input needed make to a relevant predictive attempt which makes accurate predicting available.
Machine learning enabled team performance analysis in the dynamical environment of soccer
IEEE Access
Team sports can be viewed as dynamical systems unfolding in time and thus require tools and approaches congruent to the analysis of dynamical systems. The analysis of the pattern-forming dynamics of player interactions can uncover the clues to underlying tactical behaviour. This study aims to propose quantitative measures of a team's performance derived only using player interactions. Concretely, we segment the data into events ending with a goal attempt, that is, "Shot". Using the acquired sequences of events, we develop a coarse-grain activity model representing a player-to-player interaction network. We derive measures based on information theory and total interaction activity, to demonstrate an association with an attempt to score. In addition, we developed a novel machine learning approach to predict the likelihood of a team making an attempt to score during a segment of the match. Our developed prediction models showed an overall accuracy of 75.2% in predicting the correct segmental outcome from 13 matches in our dataset. The overall predicted winner of a match correlated with the true match outcome in 66.6% of the matches that ended in a result. Furthermore, the algorithm was evaluated on the largest available open collection of soccer logs. The algorithm showed an accuracy of 0.84 in the classification of the 42, 860 segments from 1, 941 matches and correctly predicted the match outcome in 81.9% of matches that ended in a result. The proposed measures of performance offer an insight into the underlying performance characteristics.
A Survey on Football Player Performance and Value Estimation Using Machine Learning Techniques
DergiPark (Istanbul University), 2022
The popularity of games like FIFA and Football Manager has attracted millions of players. Data plays an increasingly prominent role in these games. In other words, these games contain data from all soccer players worldwide, which can be used to simulate soccer games. Therefore, a lot of experts and a lot of tools are involved in producing football data. It is not only the production of this data that has attracted researchers, but also the analysis of it since useful and consistent information can be extracted from it to design real-world football applications. In this study, we introduce a survey of studies which focus on the prediction of player value/performance using machine learning techniques. As far as we know, there is no survey in the literature that specifically addresses this topic.
A review of machine learning applications in soccer with an emphasis on injury risk
Biology of Sport, 2023
This narrative review paper aimed to discuss the literature on machine learning applications in soccer with an emphasis on injury risk assessment. A secondary aim was to provide practical tips for the health and performance staff in soccer clubs on how machine learning can provide a competitive advantage. Performance analysis is the area with the majority of research so far. Other domains of soccer science and medicine with machine learning use are injury risk assessment, players' workload and wellness monitoring, movement analysis, players' career trajectory, club performance, and match attendance. Regarding injuries, which is a hot topic, machine learning does not seem to have a high predictive ability at the moment (models specificity ranged from 74.2%-97.7%. sensitivity from 15.2%-55.6% with area under the curve of 0.66-0.83). It seems, though, that machine learning can help to identify the early signs of elevated risk for a musculoskeletal injury. Future research should account for musculoskeletal injuries' dynamic nature for machine learning to provide more meaningful results for practitioners in soccer.
Unlocking the potential of big data to support tactical performance analysis in professional soccer: A systematic review
European Journal of Sport Science
In professional soccer, increasing amounts of data are collected that harness great potential when it comes to analysing tactical behaviour. Unlocking this potential is difficult as big data challenges the data management and analytics methods commonly employed in sports. By joining forces with computer science, solutions to these challenges could be achieved, helping sports science to find new insights, as is happening in other scientific domains. We aim to bring multiple domains together in the context of analysing tactical behaviour in soccer using position tracking data. A systematic literature search for studies employing position tracking data to study tactical behaviour in soccer was conducted in seven electronic databases, resulting in 2338 identified studies and finally the inclusion of 73 papers. Each domain clearly contributes to the analysis of tactical behaviour, albeit insometimes radicallydifferent ways. Accordingly, we present a multidisciplinary framework where each domain's contributions to feature construction, modelling and interpretation can be situated. We discuss a set of key challenges concerning the data analytics process, specifically feature construction, spatial and temporal aggregation. Moreover, we discuss how these challenges could be resolved through multidisciplinary collaboration, which is pivotal in unlocking the potential of position tracking data in sports analytics.
Machine Learning Models Reveal Key Performance Metrics of Football Players to Win Matches in Qatar Stars League
IEEE Access
As football (soccer) is one of the most popular sports worldwide, winning football matches is becoming an essential aspect of football clubs. In this study, we analyzed football players' performance in a total of 864 football matches of the Qatar Stars League (QSL) between the years 2012 and 2019. For each match, the collective performance of the players in key playing positions was analyzed to understand their effectiveness in winning games. We formulated this study as a classification framework in the machine learning (ML) context to distinguish the winning team from the losing team in a match. This allowed us to check the effectiveness of different performance metrics considered a feature vector for ML models. Different ML models were considered for this classification task, and the logistic regression-based model was considered the best performing model, with more than 80% accuracy. Multiple feature selection methods were leveraged to identify players' performance metrics that could be considered as contributing factors to determine the match result. The proposed ML model identified several features, including (a) shots on target by forwarders (b) distance covered by forwarders and midfielders at very high speed (c) successful passes, that can be considered as effective performance metrics for winning a football match in QSL. Interestingly, we revealed that the defenders' role could not be ignored for match results, and playing fair games improves the chance of winning matches in QSL. We also showed that players' performance metrics from the last five seasons would provide sufficient discriminative power to the proposed ML model to predict the match-winner in the upcoming season. The proposed ML model will support the players, coaching staff, and team management to focus on specific performance metrics that may lead to winning a match in QSL.
The genesis of data-driven decision-making in the world of soccer tactics : deciphering the potential of big data
2021
Master's thesis in Information systems (IS501)Arguably, soccer is a more fluid sport than other popular sports such as baseball, American football, and cricket, which explicitly require more discrete ‘plays.’ Moreover, with twenty-two players always committed, it has more moving elements than basketball or ice hockey. From a mathematical perspective, this signifies that soccer has more degrees of freedom than these other sports, making it difficult to evaluate the game using one or a small number of metrics. Over time, this challenge has been made exceedingly more difficult within the soccer community. The availability of data has increased much more rapidly than the scientific advancements required to valorize these data. In the aftermath, most recent research papers have elaborated on specific in-game data separately and how to approach them -not how soccer clubs can assemble and utilize them through big data analytics. In accordance, an exploratory case study conducted in col...
Sports Analytics for Football League Table and Player Performance Prediction
2020 11th International Conference on Information, Intelligence, Systems and Applications (IISA, 2020
Common Machine Learning applications in sports analytics relate to player injury prediction and prevention, potential skill or market value evaluation, as well as team or player performance prediction. This paper focuses on football. Its scope is long-term team and player performance prediction. A reliable prediction of the final league table for certain leagues is presented, using past data and advanced statistics. Other predictions for team performance included refer to whether a team is going to have a better season than the last one. Furthermore, we approach detection and recording of personal skills and statistical categories that separate an excellent from an average central defender. Experimental results range between encouraging to remarkable, especially given that predictions were based on data available at the beginning of the season.
FOOTBALL ANALYTICS: A LITERATURE ANALYSIS
FROM 2010 TO 2020
Leonardo Mendes Serra Fontanive
Dissertation presented as the partial requirement for
obtaining a Master's degree in Information Management
2
NOVA Information Management School
Instituto Superior de Estatística e Gestão de Informação
Universidade Nova de Lisboa
FOOTBALL ANALYTICS: A LITERATURE ANALYSIS FROM 2010 TO
2020
by
Leonardo Mendes Serra Fontanive
Dissertation presented as the partial requirement for obtaining a Master's degree in Information
Management, Specialization in in Knowledge Management and Business Intelligence
Advisor: Mauro Castelli
January 2021
2
ABSTRACT
The overall goal for the current study is to present a literature review of analytics, precisely machine
learning (ML) reference authors in terms of methods and applicable scopes of study, in football
where is a field that historically there are empirical decisions and the usage of analytics has been
growing intensely. The research aims to list relevant academic contributions published between 2010
and 2020, performing a comparable picture per authors across the following subsets: player
individual technical skills and team performance. Furthermore, the approach will provide a summary
of studies for machine learning methods applied in football.
Such outcomes of this study would contribute to the discussion about football analytics. Regarding
that these summaries can drive researchers to have a deep dive into the fields of interest straight to
references preview studied in the thesis. Results indicate that football analytics has broadly vast
opportunities in terms of research, regarding machine learning methods and a high potential to have
a deep exploration of team and player perspective. This study can leverage and pavement new
further in-depth and targeted investigation toward football analytics.
KEYWORDS
Machine Learning; Football; Sports; Analytics; Knowledge
3
INDEX
1. INTRODUCTION ............................................................................................................ 8
1.1. Background and problem identification................................................................ 8
1.2. Problem (Research question)/ General Objective (main goal) ............................. 9
1.3. Specific Objectives ................................................................................................. 9
1.4. assumptions........................................................................................................... 9
1.5. Study relevance and importance ........................................................................ 10
1.6. Methodology ....................................................................................................... 10
2. LITERATURE REVIEW................................................................................................... 11
2.1. Football ................................................................................................................ 11
2.1.1. Overall Concept ............................................................................................ 11
2.1.2. Player individual technical skills: the complexity, samples, and limitations 12
2.1.3. Team Performance and the associated challenge ....................................... 15
2.1.4. Other Fields .................................................................................................. 16
2.2. football analytics ................................................................................................. 17
2.2.1. Concept......................................................................................................... 17
2.2.2. Use Cases ...................................................................................................... 18
3. MACHINE LEARNING MODELS .................................................................................... 28
4. AUTHORS PER FIELD: PLAYER AND TEAM .................................................................. 37
5. CONCLUSIONS ............................................................................................................ 46
5.1. Synthesis of the developed work ........................................................................ 46
5.2. Future work ......................................................................................................... 46
6. BIBLIOGRAPHY ............................................................................................................ 48
4
LIST OF FIGURES
Figure 1 - Technical performance and soccer player ratings (Pappalardo, Cintia, Pedreschi,
Giannotti, & Barabási, 2017) ............................................................................................ 12
Figure 2 – Importance of technical and contextual features to human rating process
(Pappalardo, Cintia, Pedreschi, Giannotti, & Barabási, 2017) ......................................... 13
Figure 3 - Proposed definitions for skill-related performance in soccer based on a literature
search (Aquino, Alves, Fuini, & Garganta, 2017) ............................................................. 14
Figure 4 – A sports analytics framework (Morgulev, Azar, & Lidor, 2018) .............................. 17
Figure 5 – Model of value creation from business intelligence & Analytics in competitive
sports (Caya & Bourdon, 2016) ........................................................................................ 18
Figure 6 - Screen capture from Match Vision Studio Premium® software with the categorical
matrix used in the study (Borges, Garganta, Guilherme, & Jaime, 2019) ....................... 19
Figure 7 - TeamSense software application developed in Matlab with entry point view; an
example of a time-plot variable output (e.g. surface area); an exemplar of photogram
from a variable 2D video animation (Frias, 2012) ............................................................ 20
Figure 8 - statistical indicators for improving the soccer performance from an individual’s
perspective (Vanoye, Penna, Parra, & Díaz, 2017) .......................................................... 21
Figure 9 - statistical indicators for improving the soccer performance from the team’s
perspective (Vanoye, Penna, Parra, & Díaz, 2017) .......................................................... 22
Figure 10 - Schema of the PlayeRank framework. Starting from a database of soccer-logs (a),
it consists of three main phases. The learning phase (c) is an “offline” procedure: It must
be executed at least once before the other phases since it generates information used
in the other two phases, but then it can be updated separately. The rating (b) and the
ranking phases (d) are online procedures, i.e., they are executed every time a new
match is available in the database of soccer-logs (Pappalardo, Ferragina, Cintia, &
Pedreschi, 2019) ............................................................................................................... 23
Figure 11 - Zone of ball recovered (Santos, et al., 2016) ......................................................... 25
Figure 12 - Zone of the last pass to shot (Santos, et al., 2016) ................................................ 25
Figure 13 - Shot zone (Santos, et al., 2016) ............................................................................. 26
Figure 14 - distribution of positions per event type (Pappalardo, et al., 2019) ...................... 27
Figure 15 - Player passing networks (Pappalardo, et al., 2019) ............................................... 27
Figure 16 - Model used for comparison (Khan & Kirubanand, 2019) ...................................... 34
5
Figure 17 - Left: Illustration of the football game. Right: Strategies of the hand-crafted rulebased agent (Boyd-Graber, He, Kwok, & Daumé, 2016).................................................. 34
Figure 18 - The attack leading up to Barcelona’s final goal in their 3-0 win against Real
Madrid on December 23, 2017 (Decroos, Haaren, Bransen, & Davis, 2019) .................. 41
Figure 19 – Value per action (Decroos, Haaren, Bransen, & Davis, 2019) .............................. 42
Figure 20 - The skill requirements (Key Performance Indicators) for the different positions in
soccer (Hughes, et al., 2012) ........................................................................................... 42
Figure 21 - Example of a passing event as recorded by OPTA (Kröckel, 2019) ....................... 43
Figure 22 -England’s offensive sequences ending in a shooting attempt (Kröckel, 2019) ..... 44
Figure 23 - An example of ball possession chain data. The table shows a part of a ball
possession chain dataset, which represents events in the 1st half of a match (Kusmakar
S., et al., 2016) .................................................................................................................. 44
Figure 24 - Validation of the proposed approach on the largest open collection of soccer logs
from 7 major competitions (Kusmakar S., et al., 2020) .................................................. 45
6
LIST OF TABLES
Table 1 - key performance indicators (Pratas, Volossovitch, & Carita, 2018) ........................ 15
Table 2 - For each competition, described the corresponding geographic area, the total
number of seasons, matches, events, and players. (*) 21,361 indicates the number of
distinct players, as some players play with their teams in both national and
continental/international competitions (Pappalardo, Ferragina, Cintia, & Pedreschi,
2019)................................................................................................................................. 24
Table 3 - ML Summary ............................................................................................................. 31
Table 4 - Algorithms performance on match data (number of attributes=336) (Kumar, 2013)
.......................................................................................................................................... 32
Table 5 - Algorithms performance with selected attributes (threshold=0.07, number .......... 32
Table 6 - Top 34 highest gain- ratio team attributes (Kumar, 2013) ...................................... 33
Table 7 - Evaluation of logo based replay using SVM and NN (Zawbaa, Hassanien, & ElBendary, 2011) ................................................................................................................. 35
Table 8 - Confusion matrix for event detection and summarization (Zawbaa, Hassanien, & ElBendary, 2011) ................................................................................................................. 35
Table 9 - An example of ball possession chain data. The table shows a part of a ball
possession chain dataset, which represents events in the 1st half of a match (Kusmakar,
et al., 2016)....................................................................................................................... 35
Table 10 - The predictive performance of the developed machine learning models. Shown
are the segments predicted in favor of a team with the overall prediction accuracy, the
predicted winner, and the true match results (Kusmakar, et al., 2016) ......................... 36
Table 11 – Authors Summary per field: Player and Team ....................................................... 40
7
1. INTRODUCTION
1.1.
BACKGROUND AND PROBLEM IDENTIFICATION
Football has been improving the data-driven approach in terms of the decision. Following this trend,
data researches have been emerging in the clubs and the interest in data is growing. Also, with the
increase of the Internet of Everything Technologies in Football, new sources of data are leveraging
the opportunities of analysis.
Combining data, technology, and football, (Américo, 2013) points out innovation introduced in
football led to the creation of game analysis systems. Thus, these systems analyze players' moves
collectively and individually to obtain statistics that support coaches' plans during the games.
Fundamentally, the mentioned approach uses sophisticated techniques and algorithms to deal with
important information about broad aspects of the game, such as studies of the aptitude and tactical
capacity of the players and teams present in the game.
According to (De Silva, et al., 2018), it´s possible to apply data analytics from some specific football
perspective. Also, it would be a key success to balance player indicators. Thus, for instance, tracking
in football would enable monitoring and, consequently, would be possible to determine if an athlete
is performing accordingly on matches and training sessions.
(Pires & Santos, 2018) presents a study which states that increasingly sports results often lie in the
details that can be noticed with the use of technology or device that can make the difference. Also,
nowadays in the big data era, sports are part of that, because, regarding there are large amounts of
data collected that can be applied for analysis, and sports technology is in constant growth and
development, as it has been seen the increase in science in the sports area.
Joining within the above findings, (Memmert, et al., 2011) concludes that many experts regard tactics
as the factor which gets the least attention in the training process and the work is heavily empirical
on those deficits not using a scientifically based analysis of football-specific group tactics.
Exploring the scenario presented in this study broadly, the study of (Mohammadi & Sorour, 2018)
and (Rein & Memmert, 2016) illustrates also the effects on jobs and the workforce in general when
the subject is Machine Learning, regarding that knowledge work automation addresses creative
problem-solving. Furthermore, ML techniques are enablers for big data analytics and knowledge
extraction.
(Novatchkov & Baca, 2013) summarizes, regarding gathering data in sports that it is possible to have
an insight into significant characteristics. In one way these inputs can be applied for the development
of intelligent methods adapted from conventional machine learning concepts, allowing an automatic
assessment, and providing appropriate feedback. Thus, in practice, the implementation of such
techniques could be crucial for the investigation of the quality of the execution, the assistance of
athletes but also coaches.
Analyzing the researched scenario, in the literature, there is an expressive group of machine learning
methods for football analytics. Also, among all the perspectives that involve the game, the context of
8
player individual technical skills and team performance seems to be findings that can have
interesting results, regarding the lack of scientific decision support approach.
1.2.
PROBLEM (RESEARCH QUESTION)/ GENERAL OBJECTIVE (MAIN GOAL)
Data analytics is like a storm that is gathering in football. It is one that will wash away all certainties
and change the game we know. Football analytics has not progressed as far as it could. Every shared
knowledge will support us to love more the game. Analytics in football is the newest frontier. It´s not
just a matter of collecting data. You have to know what to do with them (Anderson & Sally, 2013).
Also, technology has been changing the data collection process in high-performance sport. So it
increases the successful athletes and teams will be those who are supported by a strong analytical
capability to create all the required analysis to create competitive advantage (Schulenkorf & Frawley,
2017).
Figuring out the background and problem identification, the research aims to present relevant
academic contributions published between 2010 and 2020, performing a comparable picture per
authors across the following subsets: player individual technical skills and team performance.
Furthermore, the approach will provide a summary of studies for machine learning methods applied
in football. It´s expected to reach this general objective, giving summaries and a basis for discussion
toward football analytics in terms of methods and the referred subsets per authors.
1.3.
SPECIFIC OBJECTIVES
We can divide the primary goal of this project into the following objectives, respecting the order of
steps:
1. To investigate and study relevant academic contributions published between 2010 and 2020
toward machine learning methods applied for football analytics and the respective used fields:
player individual technical skills and team performance;
2. To evaluate and discuss the most relevant ML models per authors to perform a summary;
3. To evaluate and discuss the most relevant authors per chosen fields to perform a summary;
4. To give future directions for the next steps using this study.
1.4.
ASSUMPTIONS
The present research will underline references to build summaries about the subject of football
analytics. The proposal is to target the relevant authors and drive a literature review, not a
comparison between algorithms, regarding the accuracy and the weight of each subset approached
in the football context.
9
1.5.
STUDY RELEVANCE AND IMPORTANCE
Regarding relevance and importance, it can be highlighted that coaches can also take advantage of
the data extracted from the players’ performance. They are moving toward integrating analytics
more dynamically. Decisions may use technology to support and justify them. Coaches can make
tactical decisions, helping scouts decide what players their team should recruit as well. Coaches have
access to live analytics of the players’ physical condition and overall performance made accessible by
the analytics department that is also working in real-time in the stands (Korte, 2014).
Joining the idea, analytics has influenced the tactics in professional baseball and basketball in recent
years. Ultimately, it may have just as great an impact on football, which traditionally hasn’t relied on
statistics to figure out much of anything. In 2019, Liverpool, a recognized team in England that invest
a lot in Analytics used data to big decisions and supported them to arrive at the last UEFA Champions
League final (Schoenfeld, 2019).
Regarding the economic impact, the market for sporting events is worth $80 billion in 2014—with
impressive growth projected for the foreseeable future. For a content industry wracked with
uncertainty, sports analytics is a beacon of hope. On a sport-by-sport basis, growth occurred nearly
across the board, but football remains the runaway leader. Football revenues increased from $25.1
billion in 2009 to $35.3 billion in 2013. (Collignon, 2019).
The work of (Morgulev, Azar, & Lidor, 2018) provides an introduction that underlines big data
characteristics of sport as a uniquely authentic arena for exploring research ideas, an excellent
source for analytics and specifically those concerning certain contexts of human behavior.
Additionally, the study of (Kumar, 2013), presents that if we search at the performance analytics
literature related to football, we realize that most of the research is being done with few
performance variables and is dependent on understanding the structure of the game. The possible
interaction of multiple factors impacts the complexity of football match analysis. Thus, it makes it
necessary to improve the contribution in terms of studies and cases.
1.6.
METHODOLOGY
The design of the research will be descriptive. The first target of this research will have a
consolidated literature review about football analytics. The second step will evaluate and discuss the
most relevant ML models per authors to perform a summary, regarding the literature analysis. The
third step will be a set of relevant authors per the following fields: player individual technical skills
and team performance. At the end present the results and discussions with a conclusion and
recommendation for further researches in the future.
10
2. LITERATURE REVIEW
2.1.
FOOTBALL
2.1.1. Overall Concept
(Ali, 2011) approaches football as the premier audience sport in the world, due to its growing
popularity, as well as the amount of financial interest in the game, it is one of the most extensively
researched intermittent team sports. Highlighting, there are plenty of subject areas that have taken
advantage of scientific knowledge gained from football including the natural and physical sciences,
medicine, and social sciences.
Another important aspect to add on a timeline perspective until now, (Doidge, et al., 2019) shows
that football has been facing an economic transformation over this time and there has been a
significant change in how football clubs are managed, within an increasing focus on commercial and
media growth. Also, it has been performing investments in stadiums and branding. Furthermore,
sponsorship and contracts have been increasing exponentially to achieve global audience
penetration. For instance, the Premier League and Champions League were formed in 1992 and
these have acted as economic models that leverage the other leagues and turned it into a successful
model and push the economic transformation.
Adding new components to conceptualize football, regarding (FIFA, 2019), the world governing body,
since its birth, football has been part of our communities. It is more than a game, more than a sport,
it is a way of life that we all embrace, regardless of nationality, creed, ethnicity, education, gender, or
religion. And It is about supporting the growth and development of football by promoting the
integrity and quality of the game for everyone, and not just for today, but for generations to come.
According to (Ali, 2011) and raising the bridge to arrive on the study subject, football is a complex
sport, requiring the repetition of many disparate actions. For instance, there are several proofs of
concepts that are currently being used such as assess the physical prowess of players, approaching
simple running tests using monitor speed, agility proofs, and repeated sprint performance.
Joining what (Ali, 2011) concludes, (De Silva, et al., 2018) presents that performance management of
top football players is a complex system involving enhancement of physical performance, skill-based
training, tactical training, minimization of injury risk, and psychological support. Managing practice is
vital to allowing players to perform at an optimal level throughout a play season's length.
(Constantinou & Fenton, 2017) agrees with (Ali, 2011) and (FIFA, 2019) that Football is the most
popular sport in the world and it leverages the inspiration of several researchers to use football
activities as a real-world application field to test various statistical, probabilistic, and machine
learning techniques.
Continuing to explore the complexness, referring (Qing, et al., 2020), football is influenced by many
factors such as technical, tactical, mental, and physiological, however as the matches have a highlevel complexity and dynamic behavior, some other aspects needed to be addressed as situational
variables as match location, team quality, quality of opposition and match outcome. Also, it´s
important to underline other game nature dynamics as interactions between players and positions.
11
2.1.2. Player individual technical skills: the complexity, samples, and limitations
The football player's technical skills are directly connected with data and performance, (Caya &
Bourdon, 2016) proposes that Business Intelligence & Analytics techniques can represent individual
athletes well in their pursuit of positive achievements. Also, individual athletes are anxious to
leverage their athletic performance in their respective sport and aspire to be good at what their
sport demands in terms of physical and competitive accomplishments.
Continuing with the topic and setting practical samples, relating to football player's technical skills,
data, and performance, the study of (Spearman & Basye, 2017) presents a model for ball control in
football based on the concepts of how long it takes a player to reach the ball (time-to-intercept) and
how long it takes a player to control the ball (time-to-control). Thus, players would keep the
advantage in understanding this physics-Based modeling of pass probabilities translated and applied
to the field, regarding that these metrics are constructed at the per-player level.
Additionally, according to the figure below from (Pappalardo, Cintia, Pedreschi, Giannotti, &
Barabási, 2017), there is another sample of events produced by a player during a match which would
be considerable in terms of player individual technical skills. The study uses as a source a game
followed by reporters from three sports newspapers, then they assign an individual player rating
according to personal interpretation of each player´s performance.
Figure 1 - Technical performance and soccer player ratings (Pappalardo, Cintia, Pedreschi, Giannotti,
& Barabási, 2017)
12
Furthermore, the research of (Pappalardo, Cintia, Pedreschi, Giannotti, & Barabási, 2017) points out
one specific attention needed before analysis which is that the human evaluation process has a
limitation of features which attract their attention and then construct the evaluation. Thus, an
important step is to understand how the human evaluation process can be leveraged with the
support of data science and artificial intelligence. In the following figure, charts indicate the
importance of every attribute, normalized in the range [0; 1], to the human rating process for
football typical positions as Goalkeepers (a), Defenders (c), Midfielders (d), and Forwards (b). Hence,
taking advantage of machine learning models, as the plots indicate, most of the features have a
negligible influence on the human judge’s evaluation process.
Figure 2 – Importance of technical and contextual features to human rating process (Pappalardo,
Cintia, Pedreschi, Giannotti, & Barabási, 2017)
13
Continuing to approach the importance of context, the study of (Aquino, Alves, Fuini, & Garganta,
2017) identifies the lack of definition and classification of the skill-related variables. Also, it was
detected two additional limitations: the contextualization of the sample omitted and the influence of
match situational variables (e.g. location, quality of opponent, status); and the absence of
representative task design to measure skill-related performance. The following figure shows
definitions proposed by the research according to the literature.
Figure 3 - Proposed definitions for skill-related performance in soccer based on a literature search
(Aquino, Alves, Fuini, & Garganta, 2017)
Moreover, (Aquino, Alves, Fuini, & Garganta, 2017) focuses that a fundamental task in sports science
and performance analysis is to understand the relationship between skill acquisition and the
development of players to achieve sports excellence. Hence, it is essential to develop theoretical
principles to guide the concession of skill acquisition programs. The improvements for decisionmaking and regulation of action in dynamic environments, for instance in football, come out from the
continuous performer-environment interactions.
14
2.1.3. Team Performance and the associated challenge
Regarding the game, the last topic approached the player technical skills in terms of complexity,
samples, and limitations, there is another perspective that will be addressed in this study: the team
as a separate entity for performance analysis.
Collaborating with the statement above, the study of (Pratas, Volossovitch, & Carita, 2018) refers
that target-scoring trends have been analyzed from two different perspectives, according to research
studies performing football match analysis: the static and the dynamic. Also, the inherent
randomness of football makes the analysis even more impactful, and what makes a difference is
perhaps not the data itself, but the ability to use this data to formulate a theory that explains how a
team increases their chances of winning. Furthermore, some relevant performance metrics
correlated with goal-scoring are underlined and may be appropriate in each context of the game but
would likely be insignificant in another, depending on several factors related to the quality of the
teams and the style of the game. Thus, Football analytics determines that performance indicators are
relevant, and the problem is how the importance of different performance metrics varies depending
on the context.
Continuing to explore the challenge of analysis in terms of team performance, according to (Lepschy,
Woll, & Wäsche, 2018), despite the popularity of football and while reviews on performance
indicators in football are available, none focuses solely on the identification of success factors.
Additionally, it appeared that the most significant variables are efficiency (the number of goals
divided by the number of shots), shots on goal, ball possession, pass accuracy/successful passes as
well as the quality of opponent and match location.
Following again the research of (Pratas, Volossovitch, & Carita, 2018), the characteristics that can be
approached and analyzed on goal scoring as key performance indicators:
Pass accuracy
Number of passes
Temporal Analysis
Duration of possession
Scoring Efficiency
Number os passes
Types of passes
Game situation
Zones in which possessions started
First and next goals
Playing style
Space-time coordination
Scoring efficiency
First goal
Areas from which goals were scored
Table 1 - key performance indicators (Pratas, Volossovitch, & Carita, 2018)
The research of (Sarmento, Campanico, & Marcelino, 2014) shows an overview in terms of team
analysis, it indicates relationships between patterns of physical and efficacy of game actions
(involvements with the ball, successful, passes, dribbling, shots, and shots on target). Also, highintensity activity patterns can be a key success factor for team performance and players of more
15
successful teams covered greater total distances with the ball, and at very high-intensity running, had
a high average of goals for total shots on target, performed more actions with the ball, higher
number of passes, tackles, dribbling, and shots on target when compared with less successful teams.
2.1.4. Other Fields
Beyond the scope of this study, football has a broad scope under different fields, regarding
(Morgulev, Azar, & Lidor, 2018), sports betting market is one of largest sports business sector where
consumers and suppliers try to predict the results of future events correctly. Scientists are therefore
continuing to develop a variety of models that are constructed using different methodologies for
forecasting. Such models concentrate on predicting the outcomes of individual matches or the
outcomes of the tournament. It has become an arena for making progress in computing and machine
learning, with cutting-edge predictive analytics, due to the extremely competitive nature of the
gambling industry.
The research of (Klyuchka, Cherednichenko, Vasylenko, & Yakovleva, 2015) aimed to find the most
important factors that are not confidential information and can be easily determined before the start
of the football match. It presents that forecasting rules are used to increase the accuracy of
predicting the results of football matches by identifying the winning team based on data retrieved
from results of previous games championship, adding substantial factors, to understand the
influence of results.
The football codes are recurrent team sports with high-intensity action bursts that are intermix with
low-intensity and rest events. A variety of pressures imposed on intermittent sporting team
members contribute to temporary, acute, or chronic fatigue. Fatigue is dynamic and multifactorial
and depends on various contextual factors such as physical ability, technical abilities, the role of play,
training load, the importance of the game, and seasonal period. The number of competitive matches
per season is often very high; thus, between training sessions and competition, athletes only have a
short period to recover. There is evidence that too many matches can result in a lack of motivation
and mental burn-out, a decline in physical and match results, and an increase in injuries. There is
evidence that too many matches can result in a lack of motivation and mental burn-out, a decline in
physical and match quality, and an increase in injuries. Recovery approaches are therefore required
to relieve fatigue, recover efficiency, and reduce injury risk. (Clarke & Noon, 2019)
Such studies as (Morgulev, Azar, & Lidor, 2018), (Klyuchka, Cherednichenko, Vasylenko, & Yakovleva,
2015), and (Clarke & Noon, 2019) are outside the scope of this research, but It´s important to notice
that, conceptually, there are other fields which can be combined to cope deeply team and individual
performance, especially injury prevention according to (Clarke & Noon, 2019).
16
2.2.
FOOTBALL ANALYTICS
2.2.1. Concept
Moving forward with the football performance context in terms of team and individual player skills,
it’s important to connect the concepts of football analytics, (Babbar, 2019) shows that sports have
been facing progress from just being a sport to the involvement of science in it. The referred study
approaches Sports analytics as a combination of data collection, forecasting the game, and using
tools and techniques to interpret the game strategy to improve a player's performance individually
and for the team. Hence, it is expected that sports analysis will foster many new applications for endusers, sports coaches, and sports managers. Also, analytical goals in these applications include a
comparison of results, prediction, and behavioral correlation of attributes between players and
teams. Moreover, information can be either quantitative or qualitative and is usually collected from
the athletes' biographical data, performance, medical reports, and scouting reports.
Following the rationale of (Babbar, 2019), football analytics can support reliable and systematic data
enabling athletes and coaches to leverage their decisions. To put in place this kind of initiative, realtime systems can be used for finding key analysis points, capturing the position of ball and
movement of players throughout the game, and combining it with advanced statistical algorithms
and software would enable coaches, managers to alter their tactics to gain an upper hand on the
competitor.
According to (Morgulev, Azar, & Lidor, 2018), sports analytics can be supported as a framework,
historical data can be either quantitative or qualitative and these data are typically collected from
multiple sport-relevant resources and the collected data are standardized, centralized, integrated,
and analyzed using different metrics. Thus, it is assumed that a reliable and systematic analysis of the
data will enable different stakeholders to strengthen their decision-making processes. A sports
analytics framework is described in the following figure.
Figure 4 – A sports analytics framework (Morgulev, Azar, & Lidor, 2018)
Under this context and regard the study of (Morgulev, Azar, & Lidor, 2018), in football, teams in the
English Premier League (EPL) became advanced in terms of performance analytics. Although, when
17
compared to basketball, for instance, assessing players’ skills to score in football is slowed down by
the low frequency of scoring events. Tactical factors, such as the number and length of possessions,
passing sequences, and spatial analysis of the territory played are aggregated to optimize
performance. Additionally, a specific example of how players and coaches may benefit from the
assessment of large samples of events in football is the information combined with probability in the
directions of penalty shots, based on the shooters’ previous statistics provided by the analysts to the
goalkeepers before critical matches.
As football is a competitive sport, an integrative framework was developed by (Caya & Bourdon,
2016) in which the potential value from Business Intelligence and Analytics project is tight to the
actual focus of value where these investments are expected to happen. While this framework
provides a very high-level representation of how and where this kind of project generates value in
competitive sports, it conducts more accurate examples of value creation at each level of analysis
(institutional, organizational, and individual levels). Within each “sub-model”, specific detail about
the nature of Business Intelligence and Analytics initiatives, along with particular conversion
contingencies and measures of value created from those same initiatives.
Figure 5 – Model of value creation from business intelligence & Analytics in competitive sports (Caya
& Bourdon, 2016)
2.2.2. Use Cases
To make the concepts of football analytics tangible, this sub-section presents a compiled of samples
that show the applicability of the main purpose of the section. There are seven studies, each one
with a different perspective in the football analytics field, such as tactical efficacy, offensive
18
behaviors, defensive playing method, analysis of match situations, statistical indicators for improving
player performance, data-driven framework to evaluate player performance, situations that finished
in goal and distribution of the positions of the events.
The first sample is the study from (Borges, Garganta, Guilherme, & Jaime, 2019), it refers to tactical
efficacy and offensive game processes adopted by Italian and Brazilian youth soccer players,
approached the performance of the team with the following scope: 218 offensive actions selected
from 28 matches, including 18 matches of the Italian team U-15 in dispute for the Italian
championship, season 2015/2016, and 10 matches of the Brazilian team – 5 matches U-15 and 5
matches U-17 – in dispute for the national and state championship, season 2016. Additionally,
Matches were randomly selected along the season.
The research of (Borges, Garganta, Guilherme, & Jaime, 2019) takes advantage of data through
observational analysis using Match Vision Studio Premium®, software that enables the researcher to
create a categorical matrix according to the variables to be analyzed.
Figure 6 - Screen capture from Match Vision Studio Premium® software with the categorical matrix
used in the study (Borges, Garganta, Guilherme, & Jaime, 2019)
(Borges, Garganta, Guilherme, & Jaime, 2019) concludes that all offensive sequences ended in shots
according to the following variables: number of players involved, ball touches, passing, duration,
corridor change. Also, defines offensive actions as three: a counter-attack, quick attack, and
positional attack. Thus, with this context, the research suggests that all offensive methods adopted
can be used to achieve success during a game of U-15 and U-17 soccer players.
19
The research of (Castelão, Garganta, Afonso, José, & Costa, 2015) targets to analyze offensive
behaviors performed by six national football teams that were involved in the finals of the 2006 world
cup and 2004 and 2008 Euro Cup.
The mentioned study of (Castelão, Garganta, Afonso, José, & Costa, 2015), supported by the software
SIDS (Sequential Data Interchange Standard & GSEQ (Generalized Sequential Querier), uses
sequential analysis by the lag method to verify the different offensive game patterns and analyzes
647 offensive game sequences. Also, with the bias and sample explored, it shows that topperforming football teams drive different patterns and methods of offensive play and yet be
victorious. Furthermore, no patterns were found to be more effective than others, regarding any
specific offensive behavior.
Continuing approach football analysis on a group perspective, (Frias, 2012) presents that changes in
the defensive playing method influence the collective behavior of football teams, using Team AMS
software application and Team Sense as the support tool of the study. Moreover, it aimed to
understand the influence of specific performance constraints on the actions of teams, with the main
goal was to analyze the influence of the defensive method (zone vs. man-to-man) in the collective
performance of football teams. Then, the research analyzes two small-sided games played by two
teams of 6 players (5 outfield players plus a goalkeeper) both using zone defense in the first
experimental condition and man-to-man defense in the second one. Thus, the collective
performance of teams was captured by 4 collective variables found: surface area, stretch index,
length per with ratio, and teams’ centers' distance.
Figure 7 - TeamSense software application developed in Matlab with entry point view; an example of
a time-plot variable output (e.g. surface area); an exemplar of photogram from a variable 2D video
animation (Frias, 2012)
Regarding the study of (Frias, 2012), summarizes that the changes imposed in the collective behavior
of a team by the adoption of a different defensive method can be less strong than the own
differences between the two teams.
20
The last section approached one study that may drive a balance in terms of prioritization between
defense tactical and the opponent analysis, also, in the perspective of improving approach, the paper
from (Vanoye, Penna, Parra, & Díaz, 2017) proposes to use metrics or statistical indicators for
leveraging the football performance. Besides the rating of the individual errors with negative points:
Goals Shots Off Target, Not goals from direct free-kicks and indirect free-kicks, unsuccessful dribbles,
caught opposition offside, unsuccessful shots free-kicks or indirect free-kicks, head Shots Off the
target, shots off target, unsuccessful long /short passes, pass directions incorrectly, pass lengths
incorrect, pass locations incorrect, duels lost on the offensive/defensive, aerial duels lost on the
offensive/defensive, own goals, penalties conceded, defensive mistakes, fouls Committed, corner
crosses / direct or indirect free-kicks conceded, Throw-ins conceded, yellow or red cards, substituted
off, and others, which significantly affects the soccer performance of the team, to the metric called
Motivation Index or lack of motivation.
To going further with this Index, the study of (Vanoye, Penna, Parra, & Díaz, 2017) takes advantage of
a European football match to obtain the index of motivation and thereby determine the relationship
of the index with the outcome of the match. In the meantime, the software NacSports supports the
performance analysis, and indicators were distributed as individuals, according to figure 8, and as a
team, regarding figure 9.
Figure 8 - statistical indicators for improving the soccer performance from an individual’s perspective
(Vanoye, Penna, Parra, & Díaz, 2017)
21
Figure 9 - statistical indicators for improving the soccer performance from the team’s perspective
(Vanoye, Penna, Parra, & Díaz, 2017)
After the experimentation stage and results, (Vanoye, Penna, Parra, & Díaz, 2017) concludes that
small individual errors affect the motivation state of the players, which affects the result of the
match. Hence, indicators of motivation can be a key tool to be applied to make corrections during
the match and avoid losing the match by negative values indicators of motivation.
Among the different supporting structures oriented to analysis in football, the work of (Pappalardo,
Ferragina, Cintia, & Pedreschi, 2019) defines PlayeRank, a data-driven framework that offers a
principled multi-dimensional and role-aware evaluation of the performance of football players.
Moreover, it deployed a massive dataset of soccer-log, millions of match events on four seasons of
18 prominent soccer competitions.
The framework of (Pappalardo, Ferragina, Cintia, & Pedreschi, 2019) consists of 3 phases starting
from a database of soccer-logs: rating phase, in charge of computation of the performance rating;
ranking phase, PlayeRank assigns a player to a position according to a set of rules and based on
players rating; learning phase, generates information used in the rating and the ranking phases
performing two steps, weighting and role detector training.
22
Figure 10 - Schema of the PlayeRank framework. Starting from a database of soccer-logs (a), it
consists of three main phases. The learning phase (c) is an “offline” procedure: It must be executed
at least once before the other phases since it generates information used in the other two phases,
but then it can be updated separately. The rating (b) and the ranking phases (d) are online
procedures, i.e., they are executed every time a new match is available in the database of soccer-logs
(Pappalardo, Ferragina, Cintia, & Pedreschi, 2019)
The dataset of (Pappalardo, Ferragina, Cintia, & Pedreschi, 2019) covers a total of 64 soccer seasons,
more than 31 million events and was provided by the company Wyscout, a leading company in the
football industry that connects soccer professionals worldwide, supporting more than 50 soccer
associations, and more than 1,000 professional clubs around the world. Following the table below, all
the details will be provided.
23
Table 2 - For each competition, described the corresponding geographic area, the total number of
seasons, matches, events, and players. (*) 21,361 indicates the number of distinct players, as some
players play with their teams in both national and continental/international competitions
(Pappalardo, Ferragina, Cintia, & Pedreschi, 2019)
The research of (Pappalardo, Ferragina, Cintia, & Pedreschi, 2019) shows PlayeRank, a data-driven
framework that offers a multi-dimensional and role-aware evaluation of the performance of soccer
players which also observed that top performances are rare and unevenly distributed since a few top
players produce most of the considered excellent performances. Thus, a result that should be the
focus is that top players do not always play excellently, they just achieve top performances more
frequently than the other players. Besides, PlayeRank should be seen as a valuable tool to support
professional football scouts in evaluating, searching, ranking, and recommending soccer players.
Under team performance analysis, the research of (Santos, et al., 2016) focuses on the analysis of
match situations that finished in goal, and also 557 goals were analyzed from 10 teams across
Portugal, Spain, England, and German. Regarding those only goals that were possible to obtain the
sequence of actions from the moment of ball possession were considered. The study used the
Football Goal Observation System and found a higher number of goals happens from ball recovered
through a lost ball on the offensive zone and offensive midfield areas, where the last pass occurs in
offensive sector zones, through counterattack, within the penalty area with the right foot
corresponding 53,17% of the sample and left foot 28,83%. These results are shown in the next three
figures.
24
Figure 11 - Zone of ball recovered (Santos, et al., 2016)
Figure 12 - Zone of the last pass to shot (Santos, et al., 2016)
25
Figure 13 - Shot zone (Santos, et al., 2016)
The paper of (Pappalardo, et al., 2019) describes an open collection of soccer-logs that cover seven
male football competitions provided by Wyscout and the data were approached on the football Data
Challenge initiative (https://sobigdata-soccerchallenge.it/). Regarding that soccer-logs detailed
match events, each containing these types of information: pass, shot, foul, tackle, a time-stamp, the
player(s), the position on the field, pass accuracy, and other relevant collections.
As well as the other studies presented above, (Pappalardo, et al., 2019) explores football analytics
use cases, as the two figures described below, regarding player performance on the team context.
The first one presents the distribution of positions per event type, plotting the distribution of the
positions of the events during the match. The darker is the green, the higher is the number of events
in a specific field zone. Moreover, the same figure also describes the distribution of the passes’
position during a match for each player’s role. The darker is the color, the higher is the number of
passes in a specific field zone. Then, the second figure shows the represent player passing networks
of the match Napoli and Juventus, Italian first division, where each node is a player, and edges
represent passes between players. Moreover, The size of the nodes reflects the number of ingoing
and outgoing passes, while the size of the edges is proportional to the number of passes between the
players.
26
Figure 14 - distribution of positions per event type (Pappalardo, et al., 2019)
Figure 15 - Player passing networks (Pappalardo, et al., 2019)
27
3. MACHINE LEARNING MODELS
To evaluate and discuss the relevant ML models per authors, the study provides a summary of all the
individual studies reviewed and presented in Table 2, tracking key findings in terms of authors,
sample or dataset, methods or Algorithms, and key outputs.
There was a mix of studies, with different datasets and Algorithms. Each one has a specific approach
and achievement, but instead of evaluating isolated variables and accuracy, this chapter has the goal
to report useful methods that can bring results based on a previous investigation and may drive a
better understand of behaviors and use key reference authors as an influence to drive researches.
Starting with Kumar's research, it helps to find attributes that are relevant to assessing players'
performance and that most influence the game. Kumar explored different methods and the analysis
found supports as a guide to having a starting point both to analyze specific characteristics as well as
different machine learning algorithms that apply to football within a previously explored study.
Following table 2, the second line, the research of (Khan & Kirubanand, 2019), Besides, the
performance comparison between two algorithms such as XGBoost and SVM, has the purpose of
accuracy in terms of match prediction, this study within the current context of this work can support
potential algorithms that can be useful, concerning algorithms that behave well within the scope of
football variables.
In the sequence of the table, there is the study of (Boyd-Graber, He, Kwok, & Daumé, 2016), it
approaches the use of machine learning to take advantage of the understanding of the opponent's
behavior and adapt strategy based on predictions according to the opponent's parameters.
Within the studies presented, (Zawbaa, Hassanien, & El-Bendary, 2011) study shows the potential
use of algorithms such as support vector machine (SVM) and neural network (NN) for analysis of
football performance. The paper takes advantage of a video summarization system and shows
applicability to achieve a result of great accuracy and precision. Another preliminary analysis that can
be used as a reference for research within football.
The fifth line of the summary lists the (Kusmakar, et al., 2016) study. It analyzes the pattern-forming
dynamics of player interactions that can leverage the understanding of tactical behavior. Also, the
study explores quantitative measures of a team’s performance, focused on player interactions.
Moreover, the research shows a path that machine learning-enabled approach for automated
predictive analysis of performance.
28
Key Outputs
Methods / Algorithms
Sample/Dataset
Multilayer Perceptron
Functional Trees
59 attributes obtained Sequential Minimal
Optimization
with positive gain
The dataset of player performances for EPL released by
Naive Bayes
ratio are the ones
OPTA contained 210 attributes and 10369 instances. Out
Random Forest
affecting match
of the 210 attributes of players, 198 attributes were
Decision Table
outcome
performance statistics while the others were identifiers
Fuzzy Unordered Rule
for the player for that match. The dataset released by
Induction Algorithm
With the applied
OPTA did not contain match-outcomes. The dataset did
J48Graft
algorithms, it
not contain Own Goals, the goals scored by a player
J48
concluded top of 34
against his team. The data for Own Goals scored by each
Jrip
attributes
player in each match of the tournament was fetched from
REP Tree
characterizes the
WhoScored.com
LibSVM
match outcome to a
Kstar
satisfactory extent.
AdaBoostM1 with
Functional Tree
An ensemble learning
can be a better choice
when trying to predict
the results in this field
than SVM
SVM
Study /Author
(Kumar, 2013)
The Dataset selected contained features such as the
number of goals scored by the home team, the number of
goals scored by away team, Shots taken by the home
(Khan & Kirubanand,
team, Shots taken by away team, home team points,
2019)
away team points, a variety of betting odds, and finally
the Full-time result. The datasets collected were from the
year 2000 to 2013.
29
Key Outputs
Methods / Algorithms
DQN-world is confused by the
defensive behavior and significantly
sacrifices its performance against the
offensive opponent; DRON achieves a
much better trade-off, retaining
rewards close to both upper bounds
against the varying opponent.
DRON (Deep
Reinforcement
Opponent Network)
DQN
Game multi-agent that is played on a 6x9 grid by two
(Boyd-Graber, He,
players, which simulates movements and situations of a
Kwok, & Daumé, 2016)
football environment
Compared to the performance results
obtained using the SVM classifier, the
proposed system attained good NNbased performance results concerning
recall ratio, however, it attained poor
NN-based performance results
concerning precision ratio.
Accordingly, it has been concluded that
using the SVM classifier is more
appropriate for soccer video
summarization than the NN classifier.
ANN
SVM
Five videos for soccer matches from World Cup
Championship 2010, Africa Championship League 2010, (Zawbaa, Hassanien, &
El-Bendary, 2011)
Africa Championship League 2008, European
Championship League 2008, and Euro 2008.
Sample/Dataset
Study /Author
30
Key Outputs
Table 3 - ML Summary
machine learning-enabled approach
for automated predictive analysis of
performance and team’ s network
derived using possession chain data, by
quantitatively analyzing measures of
performance that have a specific
distribution and that can be used to
predict the performance of a team.
Methods / Algorithms
NN
SVM
Sample/Dataset
Study /Author
A dataset from a season of Major League Soccer
division of the United States and Canada. The
dataset consists of the possession chain data
from 13 matches. The interaction information
(possession chain) comprises of time and
duration of all ball passes and tackles between
players. The dataset also includes the nature of (Kusmakar, et al., 2016)
the interaction which can be categorized as being
between teammates or between opposing
players. The positional information includes the
x-y position of all individuals throughout the
entire match.
31
Regarding the study of (Kumar, 2013), it presented the following table results of algorithms
performance, sorted by correctly classified, ROC areas, F-Measure and Kappa statistic, taking into
account all the threshold levels. The top four of all the involved are multilayer Perceptron, Functional
Trees (FT), AdaboostM1 with FT, and Sequential Minimal Optimization (SMO). It shows the data used
for the classification activity is applicable and informative in terms of match outcome.
Table 4 - Algorithms performance on match data (number of attributes=336) (Kumar, 2013)
Moreover, the study of (Kumar, 2013), as moving forward on decreasing the number of attributes by
setting gain-ratio threshold the performance of some algorithms change, it concludes that the
Multilayer perceptron performance decreases as it decreases the number of attributes, but
Sequential Minimal Optimization algorithm keeps its good prediction. Also, another interesting
behavior that improved with the fewer variables approach was KSTAR.
Table 5 - Algorithms performance with selected attributes (threshold=0.07, number
of attributes=34) on match data (Kumar, 2013)
32
Besides presenting insights in terms of algorithms and the respective performance, the study of
(Kumar, 2013) brings a setlist of highest gain ratio team attributes that can support the exploration of
the relevance of football attributes. The table below is sorted by the top gain-ratio found in the
research.
Table 6 - Top 34 highest gain- ratio team attributes (Kumar, 2013)
Additionally, the algorithms approached Kumar's study, the research of (Khan & Kirubanand, 2019)
tested the performance of the highest gain-ratio team attributes applied to football. The proof
followed the figure below, regarding that the flow starts on the training and test set. The process
triggers the data clean up and the features are computed to the current data set. Moving forward
the flow, selected features are put into the SVM model with the RBF kernel. Finally, the data is fed
into the XGBoost model and the outcome from both the methods are compared.
33
Figure 16 - Model used for comparison (Khan & Kirubanand, 2019)
Extrapolating the team attribute analysis, the study of (Boyd-Graber, He, Kwok, & Daumé, 2016)
explores the concept of simulate movements and situations in a football environment, taking
advantage of a multi-agent. Developing probabilistic models or parameterized strategies for specific
applications, encoding observing through a deep Q-Network (DQN) and Deep Reinforcement
Opponent Network (DRON) variations. The following figure illustrates how the game situations and
the response against different behaviors.
Figure 17 - Left: Illustration of the football game. Right: Strategies of the hand-crafted rule-based
agent (Boyd-Graber, He, Kwok, & Daumé, 2016)
So far, these listed studies in this chapter can leverage the different types of analysis and cover
together potential gaps within tested algorithms and outcomes. Furthermore, the findings and
experimentation of (Zawbaa, Hassanien, & El-Bendary, 2011), using video as a source, adopting
neural network (NN) and SVM, can underline the key events during the match.
According to (Zawbaa, Hassanien, & El-Bendary, 2011) and following the findings, table 6 describes
an overview of the comparison between SVM and NN, and, as reported, the SVM shows a better
Performance, once the precision detected was worth 20% higher than NN. Meanwhile, Table 7 shows
the precision in terms of event detection and all the results are above 89%, which means the
proposed system achieved high accuracy.
34
Table 7 - Evaluation of logo based replay using SVM and NN (Zawbaa, Hassanien, & El-Bendary, 2011)
Table 8 - Confusion matrix for event detection and summarization (Zawbaa, Hassanien, & ElBendary, 2011)
After high accuracy achieved in the last study and (Boyd-Graber, He, Kwok, & Daumé, 2016) that
shows an approach in terms of behavior and movements, there is additional research that copes with
an analysis chain during the match. The findings of (Kusmakar, et al., 2016) can unlock the value in
terms of this type of approach.
Regarding table 8, presented by (Kusmakar, et al., 2016), shows the type of events that can be
leveraged according to a sequence during the match.
Table 9 - An example of ball possession chain data. The table shows a part of a ball possession chain
dataset, which represents events in the 1st half of a match (Kusmakar, et al., 2016)
Nevertheless, predictive models developed by (Kusmakar, et al., 2016), show a mean accuracy of up
to 75% in predicting the segmental outcome of the likelihood of a team making a successful attempt
35
to score. The following table presents the segments predicted in favor of a team with the overall
prediction accuracy.
Table 10 - The predictive performance of the developed machine learning models. Shown are the
segments predicted in favor of a team with the overall prediction accuracy, the predicted winner,
and the true match results (Kusmakar, et al., 2016)
36
4. AUTHORS PER FIELD: PLAYER AND TEAM
Five different authors are reviewed and presented in table 10, tracking key findings in terms of
authors, sample, field, and key outputs. The summary is distributed on 40% focused on the player,
40% focused on the team and 20% focused on player and team.
The first study listed in table 10, from the top to the bottom, study of (Decroos, Haaren, Bransen, &
Davis, 2019) shows a framework for valuing player actions, regarding the outcome, taking into
account the context. Thus, the player contributions can be measured according to offensive and
defensive performance.
Shifting to the next row of table 10, the article of (Hughes, et al., 2012) approaches the player
perspective as well, regarding performance indicators per position and category sets that may
support the priorities of skills driven by the game needs.
On the other hand, some authors oversee both perspectives on the same study: team and player,
such as (Kröckel, 2019). The outcome of this work is a valuable deep dive among different
approaches regarding player and team performance. It's possible to extract insights on metrics and
methods. Also, it supports an extended overview of other topics that are not the focus of this
research, but it should be considered in terms of a reference to good references.
Moving forward with table 10, (Kumar, 2013) research presents a reference in terms of team
perspective. Besides the last chapter, where this author was explored, the study can leverage
specifically the team performance and attributes that may influence the game.
Last but not least, the study of (Kusmakar S., et al., 2020) shows the potential for uncovering local
numerical markers of team performance. Regarding the topic ML was driven in chapter 9, it’s
important to underline that this research also has a machine learning-enabled approach. Moreover,
it has valuable insights into team analysis.
37
Study / Author
Sample / Dataset
Field
Wyscout data for the English, Spanish, German,
Italian, French, Dutch, and Belgian top
(Decroos, Haaren, Bransen, & Davis,
divisions. It considered 11,565 games played in
2019)
the 2012/2013 through 2017/2018 seasons.
Player
European Football Championships of 2004. The
measure was based on a subjectively drawn
continuum that analyses a player’ s technical
movement throughout the game
Player
(Hughes, et al., 2012)
Key Outputs
Values all action types as
passes, crosses, dribbles, and
shots
Reasons about an action’ s
possible effects on the
subsequent actions.
The player actions that
increase and decrease a
team’ s chance of scoring
Key performance indicators
per position and category
sets
38
Study / Author
(Kröckel, 2019)
Sample / Dataset
Field
Key Outputs
Social network analysis (SNA)
Dynamic network analysis
(DNA)
Overview of data and tools
in SNA research in football
Social network metrics used
Euro 2016 dataset among 51 games from OPTA
in football performance
database, not all games are used for each
analysis
approach during the study. A selection was Player and Team
SOM network be applied for
performed based on the aim of the analysis
team performance analysis
and the required amount of data.
Comparison of mining
algorithms, regarding
football
Clustering algorithms
Data and information useful
for real-time decision
support
39
Table 11 – Authors Summary per field: Player and Team
Study / Author
Sample / Dataset
Field
(Kumar, 2013)
The dataset of player performances for EPL
released by OPTA contained 210 attributes and
10369 instances. Out of the 210 attributes of
players, 198 attributes were performance
statistics while the others were identifiers for
the player for that match. The dataset released
by OPTA did not contain match-outcomes. The
dataset did not contain Own Goals, the goals
scored by a player against his team. The data
for Own Goals scored by each player in each
match of the tournament was fetched from
WhoScored.com
Team
A dataset from a season of Major League
Soccer division of the United States and
Canada. The dataset consists of the possession
chain data from 13 matches. The interaction
information (possession chain) comprises of
time and duration of all ball passes and tackles
between players. The dataset also includes the
nature of the interaction which can be
categorized as being between teammates or
between opposing players. The positional
information includes the x-y position of all
individuals throughout the entire match.
Team
(Kusmakar S., et al., 2020)
Key Outputs
Top of 34 team attributes
that affect the match
outcome
Potential for uncovering
local numerical markers of
team performance.
40
Moving back to (Decroos, Haaren, Bransen, & Davis, 2019) research, the framework studied has
different ratings, regarding actions of players’ offensive and events influence. As an example of this
kind of approach, there is the figure below.
.
Figure 18 - The attack leading up to Barcelona’s final goal in their 3-0 win against Real Madrid on
December 23, 2017 (Decroos, Haaren, Bransen, & Davis, 2019)
Following the research of (Decroos, Haaren, Bransen, & Davis, 2019), figure 18 illustrates exactly
what is stated in the title of the article: actions speak louder than goals. And connecting the next
figure 19 shows a correlation between the number of actions per 90 minutes with the related value
per action, the result of the finding demonstrates that Lionel Messi is a world-class player.
41
Figure 19 – Value per action (Decroos, Haaren, Bransen, & Davis, 2019)
Keeping in mind the player perspective, the study of (Hughes, et al., 2012) presents the following
figure, indicating key performance Indicators (KPI) that best fit per position. Thus, it’s a valuable
reference to have a deep dive and tackle the important KPI regarding the player's objective during a
match.
Figure 20 - The skill requirements (Key Performance Indicators) for the different positions in soccer
(Hughes, et al., 2012)
42
Shifting to another author, the research of (Kröckel, 2019) approaches some relevant topics among
football analytics, on the player and team perspective. The results of the study present insights on
metrics and methods, catching how coaches can take advantage of decision support during live
matches.
One way to structure data, regarding (Kröckel, 2019), is to track a single event., for instance, shot on
goal, in a manner that it’s possible to define the actor (who), the type of action (what), time (when)
and the pitch position (where). The figure below shows the file format that aggregates this kind of
information.
Figure 21 - Example of a passing event as recorded by OPTA (Kröckel, 2019)
Continuing on the same study, it shows a long list of authors and depicts player performance analysis
and team performance, in a specific chapter, largely exploring social network analysis and dynamic
network analysis.
Another interesting highlight on the (Kröckel, 2019) research is the chapter that deeply dives on the
tactical behavior and the individual actions performed, regarding mining algorithms. Also, it explores
a sequence of game steps ending on a final action target, as demonstrated in the following figure.
43
Figure 22 -England’s offensive sequences ending in a shooting attempt (Kröckel, 2019)
Changing and looking forward to a reference about team perspective, there is the research of
(Kumar, 2013), mentioned also on the last chapter, also (Kusmakar S., et al., 2020) analyzes football
as a dynamic system, regarding player actions and the context among events as it demonstrates on
the following figure.
Figure 23 - An example of ball possession chain data. The table shows a part of a ball possession
chain dataset, which represents events in the 1st half of a match (Kusmakar S., et al., 2016)
Keeping on the same study of (Kusmakar S., et al., 2016), another key finding on it, is the
combination of machine learning with team performance, achieving an automated machine learning
model to predict the outcome of a game segment, comparing another authors and researches as
illustrated on the next figure. It can drive results between different approaches.
44
Figure 24 - Validation of the proposed approach on the largest open collection of soccer logs from 7
major competitions (Kusmakar S., et al., 2020)
45
5. CONCLUSIONS
5.1.
SYNTHESIS OF THE DEVELOPED WORK
The objective of this study was to investigate and explore relevant academic contributions toward
machine learning methods applied for football analytics and the respective used fields were the
following: player individual technical skills and team performance. Furthermore, assess the most
relevant ML models per authors on the football analytics field to deliver a summary. Besides, the aim
was to evaluate the most relevant authors per chosen fields to deliver a summary as well. Finally, in
this final chapter, give future directions for the next steps using this study.
In the first part of the research, the context of football analytics background was approached on the
literature review. The boundaries and fundamentals of the study were established to drive the path
for the next steps.
After that, analyzing the literature, the summary of machine learning methods among football
analytics had five relevant findings which had a concentration on SVM and different kinds of neural
network algorithms.
After comparing machine learning methods, the analysis went down on authors per the defined
boundaries of scope perspective: team and player. Regarding results, there were five satisfactory
studies in which it’s possible to explore valuable player and team metrics, the importance of actions,
and the sequence itself. Thus, the applicability of football analytics-driven on the specific fields
approached on this research.
Despite the sample size found in the study, the results from the literature can drive other researchers
to have a guide and, through compiled summaries, pave new further in-depth and targeted
investigation toward football analytics.
Besides, it was concluded that football analytics has broadly vast opportunities in terms of research,
regarding machine learning methods and a high potential to have a deep exploration of team and
player perspective.
5.2.
FUTURE WORK
As the next step, I would explore deeply samples and experiments in the field of machine learning
application, as long as use cases taking advantage of team and player as the target. Nevertheless, it
would be great to apply in a real-life world and follow up results, supporting football decisions and
tracking achievements.
Second, future research might also further investigate the usage of the SVM, regarding that this
model was found in most of the studies showed in the present research, and compare the results
obtained to other machine learning algorithms, applying football datasets and targeting a player and
teams analysis, identifying reliably and accurately.
Last but not least, football has alternative perspectives that influence team results and should be
explored, regarding player recovery in terms of fatigue, psychological mindset, and other relevant
aspects that may be addressed. Thus, another further investigation would be the correlation
46
between team results and other potential perspectives taking advantage of previous studies, and the
list of researches already touched on the presented work.
47
6. BIBLIOGRAPHY
Américo, J. (2013). Sistema de Seguimento de Jogadores de Futebol baseado em Vídeo de Baixa
Qualidade. Porto: Faculdade de Ciências da Universidade do Porto em Ciência de
Computadores.
Anderson, C., & Sally, D. (2013). The Numbers Game: Why Everything You Know About Soccer Is
Wrong. London: Penguin Group.
Aquino, R., Alves, I., Fuini, E., & Garganta, J. (2017, June). Skill-Related Performance in Soccer: A
Systematic Review. Human Movement.
Babbar, M. (2019). A systematic review of sports analytics.
Borges, P., Garganta, J., Guilherme, J., & Jaime, M. (2019, August). Tactical efficacy and offensive
game processes adopted by Italian and Brazilian youth soccer players. Motriz. Revista de
Educação Física.
Boyd-Graber, J., He, H., Kwok, K., & Daumé, H. I. (2016). Opponent Modeling in Deep Reinforcement
Learning. Proceedings of the 33rd International Conference on Machine. New York.
Castelão, D., Garganta, J., Afonso, José, & Costa, I. (2015, June). Sequential analytsis of attacking
bahaviors performed by top-level national soccer teams. Revista brasileira de ciência do
esporte, pp. 230-236.
Caya, O., & Bourdon, A. (2016). A framework of value creation from business intelligence and
analytics in competitive sports. 49th Hawaii International Conference on System Sciences
(HICSS) (pp. 1061-1071). IEEE.
Clarke, N., & Noon, M. (2019). Editorial: Fatigue and Recovery in Football. Coventry University.
Collignon, H. (2019). Winning in the Business of Sports. Retrieved from Atkearney:
https://www.atkearney.com/communications-media-technology/article?/a/winning-in-thebusiness-of-sports
Constantinou, A., & Fenton, N. (2017). Towards Smart-Data: Improving predictive accuracy in a longterm football team. Knowledge-Based Systems.
De Silva, V., Caine, M., Skinner, J., Dogan, S., Kondoz, A., Peter, T., . . . Smith, B. (2018, October 26).
Player Tracking Data Analytics as a Tool for Physical Performance Management in Football: A
Case Study from Chelsea Football Club Academy.
Decroos, T., Haaren, J. V., Bransen, L., & Davis, J. (2019, August). Actions Speak Louder than Goals:
Valuing Player Actions in Soccer. The 25th ACM SIGKDD Conference on Knowledge Discovery
and Data Mining (KDD ’19).
Doidge, M., Claus, R., Gabler, J., Irving, R., Millward, P., & Silvério, J. (2019). The impact of
international football events on local, national, and transnational fan cultures: a critical
overview. Soccer & Society.
48
FIFA. (2019, September 29). FIFA. Retrieved from https://www.fifa.com/livingfootball
Frias, T. (2012). Changes in defensive playing methods influence the collective behavior of association
football teams. Universidade Técnica de Lisboa.
Hughes, M., Caudrelier, T., James, N., Redwood-brown, A., Donnelly, I., Kirkbride, A., & Duschesne, C.
(2012). Moneyball and soccer - An analysis of the key performance indicators of elite male
soccer players by position. Journal of Human Sport and Exercise, 7(2).
Khan, S., & Kirubanand, V. (2019). Comparing machine learning and ensemble learning in the field of
football. International Journal of Electrical and Computer Engineering (IJECE), 4321-4325.
Klyuchka, Y. A., Cherednichenko, O. Y., Vasylenko, A. V., & Yakovleva, O. V. (2015). Forecasting the
results of football matches on the internet-based information. Bulletin of NTU "KPI".
Korte, T. (2014, June 19). Datainnovation.org. Retrieved from Datainnovation.org:
https://www.datainnovation.org/2014/06/how-data-and-analytics-have-changed-thebeautiful-game/
Kröckel, P. (2019, July). Big Data Event Analytics in Football for Tactical Decision Support.
Kumar, G. (2013). Machine Learning for Soccer Analytics. Dublin.
Kusmakar, S., Shelyag, S., Zhu, Y., Dwyer, D., Gastin, P., & Angelova, M. (2016). Machine learningenabled team performance analysis in the dynamical environment of soccer.
Kusmakar, S., Shelyag, S., Zhu, Y., Dwyer, D., Gastin, P., & Angelova, M. (2020, March). Machine
learning-enabled team performance analysis in the dynamical environment of soccer. DSI
Collaborative Research (Intelligent Sensor Processing for Enhancing Defence Decision
Support).
Lepschy, H., Woll, A., & Wäsche, H. (2018). How to be Successful in Football: A Systematic Review.
The Open Sports Sciences Journal.
Memmert, D., Bischof, J., Endler, S., Grunz, A., Schmid, M., Schmidt, A., & Perl, J. (2011). World-Level
Analysis in Top Level Football Analysis and Simulation of Football Specific Group Tactics by
Means of Adaptive Neural Networks. Artificial Neural Networks.
Mohammadi, M., & Sorour, S. (2018, June 5). Deep Learning for IoT Big Data and Streaming Analytics:
A Survey. IEEE COMMUNICATIONS SURVEYS & TUTORIALS.
Morgulev, E., Azar, O. H., & Lidor, R. (2018). Sports analytics and the big-data era. International
Journal of Data Science and Analytics.
Novatchkov, H., & Baca, A. (2013, December). Artificial Intelligence in Sports on the Example of
Weight Training. Journal of Sports Science and Medicine, pp. 27-37.
Pappalardo, L., Cintia, P., Pedreschi, D., Giannotti, F., & Barabási, A.-L. (2017, December). Human
Perception of Performance. arXiv on Physics and Society.
49
Pappalardo, L., Cintia, P., Rossi, A., Massucco, E., Ferragina, P., Pedreschi, D., & Giannotti, F. (2019). A
public data set of spatio-temporal match events in soccer competitions. Scientific Data.
Pappalardo, L., Ferragina, P., Cintia, P., & Pedreschi, D. (2019, September). Player Rank: Data-driven
Performance Evaluation and Player Ranking in Soccer via a Machine Learning Approach. ACM
Transactions on Intelligent Systems and Technology, 10.
Pires, M., & Santos, V. (2018). Assessing the Impact of the Internet of Everything Technologies in
Football. Journal of Sports Science 6, pp. 36-55.
Pratas, J. M., Volossovitch, A., & Carita, A. I. (2018). Goal scoring in elite male football: A systematic
review. Journal of Human Sport and Exercise.
Qing, Y., Ruano, M.-Á., Hongyou, L., Zhang, S., Gao, B., Wunderlich, F., & Memmert, D. (2020).
Evaluation of the Technical Performance of Football Players in the UEFA Champions League.
International Journal of Environmental Research and Public Health.
Rein, R., & Memmert, D. (2016). Big data and tactical analysis in elite soccer: future challenges and
opportunities for sports science. SpringerPlus.
Santos, F., Mendes, B., Maurício, N., Furtado, B., Sousa, P., & Pinheiro, V. (2016). Análise do Golo em
Equipas de Elite de futebol na época 2013-2014. Revista de Desporto e Actividade Física, 8,
11-22.
Sarmento, H., Campanico, J., & Marcelino, R. (2014). Match analysis in football: a systematic review.
Journal of Sports Sciences.
Schoenfeld, B. (2019, May 22). Nytimes. Retrieved from Nytimes:
https://www.nytimes.com/2019/05/22/magazine/soccer-data-liverpool.html
Schulenkorf, N., & Frawley, S. (2017). Critical Issues in Global Sport Management. London: Routledge.
Spearman, W., & Basye, A. T. (2017). Physics-Based Modeling of Pass Probabilities in Soccer. Sports
Analytics Conference. MIT Sloan.
Tax, N., & Joustra, Y. (2015). Predicting The Dutch Football Competition Using Public Data: A Machine
Learning Approach. TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, VOL. X, NO.
X, MONTH YEAR 1.
Vanoye, J., Penna, A., Parra, O., & Díaz, D. (2017). Motivation Index to Improve Soccer Performance.
International Journal of Combinatorial Optimization Problems and Informatics, 8, 45-57.
Zawbaa, H. M., Hassanien, A. E., & El-Bendary, N. (2011). Machine Learning-Based Soccer Video
Summarization System. Communications in Computer and Information Science, (pp. 19-28).
50
Page | i
Related papers
Trayectorias del color de la tierra: aproximaciones al movimiento indígena en México desde la dialogicidad artístico-política y la política prefigurativa. (Capítulo de libro).
Arte y Cultura para la vida. Cultura, políticas y resistencias en la sociedad mexicana del siglo XXI, 2024
Anticolonialism & Nationalism: Deconstructing Synonymy, Investigating Historical Processes
HAL (Le Centre pour la Communication Scientifique Directe), 2012
Inferencia estadística utilizando redes neuronales artificiales
Scientia Et Technica, 2004
Resonances: Hereke Factory Children From Empire to Republic (1890-1960)
Multiple Layers of Childhood Landscape: Emotional, Physical, and Spatial Encounters in Late Ottoman Empire and Early Republican Turkey Panel III-09, sponsored byAssociation of Middle East Children and Youth Studies, 2024 MESA Annual Meeting On Tuesday, November 12 at 11:30 am, 2024
Green Chemical Syntheses and Processes: Introduction
ACS Symposium Series, 2000
Linguistic and social typology: The Austronesian migrations and phoneme inventories
Linguistic Typology, 2004
Perencanaan Strategis Sistem Informasi dengan Menggunakan Metode Ward and Peppard pada Perusahaan Distributor
Jurnal Bina Komputer
Development of a Lagrange-Monte Carlo scheme for fluid modelling of SOL/divertor plasmas
Contributions to Plasma Physics, 2018
An Uncommon Recurrent Metastasis of Ovarian Immature Teratoma to the Small Bowel
Case Reports in Oncology, 2021
Elevated Leukocyte Glucose Index Is Associated with Long-Term Arteriovenous Fistula Failure in Dialysis Patients
Journal of clinical medicine, 2024
CANGKANG Pila ampullacea DAN BLOTONG TEBU SEBAGAI BAHAN PEMBUATAN SMART-PARTISI
Jurnal Riset Rekayasa Sipil
Sub-Micrometer Anatomical Models of the Sarcolemma of Cardiac Myocytes Based on Confocal Imaging
Biocomputing 2008 - Proceedings of the Pacific Symposium, 2007
Labor induction in term premature rupture of membranes: comparison between oxytocin and dinoprostone followed 6 hours later by oxytocin
American Journal of Obstetrics and Gynecology, 2012