International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169

Volume: 6 Issue: 2 162 - 165

______________________________________________________________________________________
Prediction of Football Match Score and Decision Making Process

Nilay Zaveri Utkarsh Shah Shubham Tiwari

IT-SAKEC IT-SAKEC IT-SAKEC
Mumbai, India. Mumbai, India. Mumbai, India.
[email protected] [email protected] [email protected]

Pramila Shinde
Assistant Professor Lalit Kumar Teli
IT-SAKEC IT-SAKEC
Mumbai, India. Mumbai, India.
[email protected] [email protected]

Abstract— Football isone of the most popular sport internationally. With that amount of popularity and fame, it has gained a lot of attention
from the technical field. Prediction of match results has been in demand for various purposes including team management, analysis and betting.
We have come up with a solution using machine learning algorithms that can fulfil all the current needs in football match prediction. The
implementation only includes teams from Spanish La Liga over the last 5 seasons. We have predicted the outcome of matches between Home
Team and Away Team which would include the final score, the starting 11 players, the substitutes and the names of probable goal scorers. For
the purpose of analysis, we have provided the stats of players and teams referring to the FIFA 18 game database as well as their actual career
stats. We have also provided the analysis of strength, weakness and tactics of players and teams. Finally, for decision making purposes, we
would make a system that can analyse the Home and Away team and then suggest the tactics to the user for their team that can maximize their
winning chances.
_________________________________________________*****_________________________________________________

I. INTRODUCTION dependent variable has 3 possible outcomes (Home Win,

Away Win and Draw) and hence, multinomial regression has
Today, the team managers,fans and analysts give a prediction
been used in a one vs all mode. This method’smain advantage
about who is going to win the match. Nowadays, different
is that it’s very simple to explain the relationship between
statistics are used for the prediction of football match result.
output variable and input. It allows one to say that the
The prediction of a football match is done by using some
presence of a risk factor increases the probability of a given
previous data. The prediction of the match depends on
outcome by a specific percentage. We proposed a model of
variables like Team stats, Player stats, Historical data
football match prediction by using the data of every match in
etc.which is used by managers and club directors to decide
La Liga for the last 5 years and training the algorithm based
who is going to win the match and what is needed to win the
on that data. The algorithm learns from training data to form
match. Football result prediction has gained lots of popularity
different rules and pattern and based on that they make a
in recent years. Previously, Knowledge discovery in databases
decision on some new data. With every input, the algorithm
(KDD) was used to develop a football match result predictive
keeps on learning by using the feedback.
model by gathering the features that affect the outcome of
football matches. Data mining techniques have also been used The software predicts the match result, score, starting 11,
in thepast.In existing systems, they have analysed two substitutes and names of goal scorers. These results can be
approaches used in football match result prediction, statistical used by the management staff while team selection or by the
and machine learning. For team management, the manager fans playing fantasy leagues in order to build an efficient team
and the other coaching staff members need to know about or for the purpose of betting. This software is different from
their own team as well as the opponent team in detail. They the other software as it uses the machine learning algorithms
should be able to identify strengths and weaknesses of along with a little statistical approach to improve the
different teams and accordingly prepare for the matches. The accuracy. Another difference is that it can be used by all types
management should also be able to monitor a match in real of users (fans, management staff and bet placers) because it
time to take a decision during a match. For all these purposes, provides a wide range of functionalities rather than just
the prediction application should be able to suggest ideas prediction of Win, Lose or Draw.
based upon the provided conditions. We have used multiple
A. AIM OF THE PROJECT
machine learning algorithms wherein Logistic Regression
gave us the best accuracy. Logistic Regression is used when The software product “PREDICTION OF FOOTBALL
the dependent variable is categorical. In our case, the MATCH SCORE &DECISION MAKING PROCESS” will be
162
IJRITCC | February 2018, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 6 Issue: 2 162 - 165
______________________________________________________________________________________
a system that provides prediction of match result, squad, as an Away Win. The factor that matches can result to a Draw
substitutes, goal scorers, team analysis, player analysis, player hasn’t been considered. [3]
evaluation and decision support to improve the winning
Predicting football scores using machine learning techniques
chances against any given team.
MIPRO 2011, MAY 23, 2011, Opatija, Croatia
B. SCOPE OF THE PROJECT
They have implemented the following algorithms:
This software will improve the efficiency and will be able to
address all the needs of the different users. It will help team  Naive Bayes
managementto know about their own team as well as the  Bayesian networks
opponent team in detail. They’ll be able to identify strengths  LogitBoost
and weaknesses of different teams and accordingly prepare for  The k-nearest neighbors algorithm
the matches. The management also is able to monitor a match  Random forest
in real time to take a decision during a match.  Artificial neural networks.
II. LITERATURE SURVEY Using those algorithms, they have predicted Home Win, Away
An Improved Prediction System for Football a Match Result Win and Draw. The best accuracy that they could reach was
65%. The drawback of the system is that it uses a very small
Vol. 04, Issue 12 (December 2014) dataset (96 matches) i.e. the data of 1 complete UCL season
They have used an approach of KDD using Logistic and the testing is done on the same. [4]
Regression and ANN. They have performed operations using A novel way to Soccer Match Prediction
RapidMiner tool. The application only contains a data of 110
matches of the 2014-2015 EPL and have predicted the results They have used multiple machine learning algorithms for the
of the further matches of the same season.The data used is prediction of Home Win, Away Win and Draw. The system
very specific to a particular season and the performance of uses the data from FIFA 15 game and also Match History
different teams, players and coaches in the starting matches database of EPL season 14-15. The training and testing is
of the season. They have been able to reach an accuracy of done on the same season database and the best accuracy
93%. The drawback of the system is that it can only predict obtained is 80%. The drawback of the system is the fact that it
Win, Lose or Draw and only for a particular season. Since, only considers 1 season for training and the testing is done on
there is always a major change when a new season begins, the the same.[5]
accuracy of the model for a fresh season can be highly
III. PROBLEM DEFINITION
affected. [1]
Prediction of the result(Home Win, Away Win or Draw) of a
Football Result Prediction with Bayesian Network in Spanish
football match in La Liga by specifying the Home and Away
League-Barcelona Team
team names. Also predicting the exact score-line of the match
Vol. 5, No. 5, October 2013 along with the names of the probable goal-scorers and also
other football match attributes like shots, shots-on-target and
They have used Bayesian Networks for classification. The
corners. Finally, include a decision suggestion system for the
system is set for only 1 team (FC Barcelona) and has a vast
football team managers to improve the team’s performance.
number of real-time parameters (Psychological and non-
The decision support system would suggest team selection,
psychological factors). Every match is considered separately
tactic selection, analysis of opponent team, player evaluation
since all the factors can change with time.The accuracy of the
etc.
model was 92%. The drawback of the system is that it can
only predict for a particular team and needs data input for IV. DESIGN & IMPLEMENTATION
each match before prediction. Also, it can only predict win,
A. Database Design
lose or draw. [2]
Five databases have been used, each for a different
Predicting Football Match Results with Logistic Regression
application:
They have used Logistic Regression algorithm for prediction of  The Match History Database for La Liga 12-13 to La
results of 1 season based on the previous 5 seasons. They are Liga 16-17 (5 years) has been used which contains 12
only predicting Home Win or Away Win. They have obtained different attributes.
an accuracy of 69.5%. The drawback of the system is that it  Also, a Team Vs Team database has been used which
only predicts the probability for Home Win. If the probability contains the records for all the teams against all
is more than 50% then it is specified as a Home Win otherwise otherteams in the last 5 seasons in La Liga. These 2 tables
163
IJRITCC | February 2018, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 6 Issue: 2 162 - 165
______________________________________________________________________________________
are together used forprediction of match result (Home implemented using a statistical approach along KDD to find
Win, Away Win or Draw). the strength and weakness of all the teams based on which
 Goals history database contain several attributes related suggestions will be made to improve the winning chances.
to goals, shots, shots on target etc. of the last 5 seasons
and is used for the prediction of final scores for each
team.
 Player Stats database has been taken from “FIFA 18”
game which is used for Player Analysis, Player
Evaluation and for selection of Starting 11 players along
with the substitutes. It includes 46 different attributes for
each player. The database also helps for prediction of
goal-scorers.
 Team Stats database has been taken from “FIFA 18”
game which will be used for Team Analysis and
Suggestion system.

B. Implementation
The system has been implemented using several machine
learning algorithms (Logistic Regression, Random Forest,
Artificial Neural Network, Linear SVM and Naïve Bayes)
using Python Sci-Kit Learn library. The training and testing is
done on the complete dataset using Cross-Validation function
with cv = 10. The main part of our implementation was
concentrated on database pre-processing and attribute
selection.
The Match History Database includes 12 attributes: Home
Team ID, Home Shots, Home Shots on Target, Home
Corners, Home Yellow Cards, Home Red Cards and similar 6
attributes for Away team. A total of 29 teams have
participated in La Liga between 2012-2017. These teams have
been assigned an ID from 1-29 based on their performance in
the 5 seasons where the best team ID starts from 1.
As the 13th attribute we used the Home Team’s win
percentage against the given Away Team, which was obtained
from the Team Vs Team Database. It proved to be successful
by increasing the accuracy of the model by almost 8%. This
idea was taken from the fact that every team has a playing Figure1. Project Flow Diagram
style and also has rivals. So, even though a team is better than
the other team in everything, but due to the rivalry or due to
C. Results Obtained
their respective playing styles, the result can be a loss. Hence,
TABLE 1. ALGORITHM ACCURACY TABLE
the inclusion of Team ID and Head to Head win percentage
LR RF ANN Linear NB
was successful. SVM
Further, we used the Goals History database to predict the Match 63.94% 61.53% 63.1% 58.25% 58.63%
exact score of both the teams at the end of the match using History
Logistic Regression and the accuracy was found out to be Database
69.87%. After this, we used the statistical and random Match 71.63% 69.9% 69.2% 66.95% 63.57%
simulation approach to find the goal scorers. The team and History
player analysis were displayed directly from the databases and Database
represented graphically as in [5]. +
Later we would make the player evaluation system to find Team Vs
whether a new player can fit in a squad and also their transfer Team
Database
value. Finally, the decision support system will be
164
IJRITCC | February 2018, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 6 Issue: 2 162 - 165
______________________________________________________________________________________
LR = Logistic Regression VI. REFERENCES
RF = Random Forest [1] Igiri,Chinwe Peace, Nwachukwu, Enoch Okechukwu,“An
ANN = Artificial Neural Network Improved Prediction Systemfor Football a Match Result”
IOSR Journal of Engineering (IOSRJEN), Vol. 04, PP 12-
NB = Naïve Bayes
20.
SVM=Support Vector Machine [2] FarzinOwramipur, ParinazEskandarian,and
FaezehSadatMozneb,“Football ResultPrediction with
The Accuracy specified in the table is the average of the Bayesian Network in Spanish LeagueBarcelona Team”
accuracy for Home Win, Away Win and Draw classifications. InternationalJournal of Computer Theory and Engineering,
Vol. 5, No. 5, October 2013.
[3] Darwin Prasetio, Dr.Harlili,“Predicting Football
D. Regression Coefficients MatchResults with LogisticRegression”, Institute Teknologi
Bandung, 2016.
TABLE 2. REGRESSION COEFFICIENTS TABLE [4] Josip Hucaljuk, AlenRakipović,“Predicting football scores
Home Win Away Win Draw using machine learningtechniques”, University of Zagreb,
HT ID -0.027 -0.008 0.020 Faculty of Electrical Engineering and Computing Unska 3,
May 23-27, 2011.
AT ID 0.009 -0.015 -0.004 [5] Jongho Shin, Robert Gasparyan, “A novel way to Soccer
HT Win 4.613 -4.836 -0.630 Match Prediction”.
percentage
vs AT
HT Shots -0.112 0.072 0.030
AT Shots 0.053 -0.089 0.024
HT Shots on 0.449 -0.275 -0.193
Target
AT Shots on -0.347 0.560 -0.193
Target
HT Corners -0.155 0.083 0.060
AT Corners 0.089 -0.117 0.016
HT Yellow -0.036 0.027 0.034
Cards
AT Yellow -0.087 0.017 0.104
Cards
HT Red -0.650 0.435 0.128
Cards
AT Red 0.489 -0.722 0.093
Cards

HT: Home Team

AT: Away Team

V. CONCLUSION
We implemented the model using different machine learning
algorithms and were able to reach the accuracy of 71.63%
with Logistic Regression on the Match History Database of 5
seasons along with the Team Vs Team Database.Referring to
related works, there are a lot of improvements that can be
made to our system for improving the accuracy of both, match
result prediction as well as goal prediction. Also, the two
predictions can be later combined to further improve the
model. We can also use the Machine Learning approach for
the Goal Scorer prediction.
165
IJRITCC | February 2018, Available @ http://www.ijritcc.org
_______________________________________________________________________________________