Turkish Journal of Computer and Mathematics Education Vol.10 No.

03 (2019), 1341-1347
Research Article

ENHANCED SLAP ALGORITHM TO FIND TOP FREQUENT ITEMSETS IN THE

ONLINE RETAIL STORE

a
V. Ravi kumar, bDr. V. Purnachandra Rao.

a
Associate Professor of Information Technology, TKREC, Hyderabad, Telangana, India
b
Professor,SVIT, Hyderabad, Telangana, India.
a
[email protected]
b
[email protected]

_____________________________________________________________________________________________________

Abstract: Online shopping plays an important role from the past few decades. E-commerce websites to increase their profits
by performing business analytics on the top utility items. Existing transactional databases implemented sequential patterns to
find top utilities. More number of resources and computations are required to access these databases. In the proposed
research, the model recognizes the frequent patterns by extending the Genetic Approach known as “Slap”, a swarm
optimization which finds the best solution for the frequent activities. Optimization technique in this research is associated
with finding the top-k utilities with maximum accuracy. The major advantage of this approach lies in its colony formation in
perfect shape by communicating with neighbours. The elements in the local search space continuously communicate with
their agents and update their positions in global space. The implementation of genetic approaches with multi objective
function helps the model to reduce the features and improve the accuracy. The proposed approach has obtained 97.1%
accuracy.

Keywords: Transaction Databases, Sequential Mining, Salp Algorithm, Swarm Intelligence, Accuracy

___________________________________________________________________________

1. Introduction
Eclat, which searches the subset tree in-depth first. The effectiveness of processing memory consumption and
time use between the old methods and the suggested algorithm was the authors' main concern. The itemset's
occurrences would be among the aspects affecting the algorithm's performance. A set of members of the class for
a sub tree rooted at P frequent items formed by computing diffset for all unique pairings of itemsets and
evaluating the support of the resulting itemsets serves as the input to the technique. With those item sets that are
discovered to be common at the current level, a recursive process call is made. Until all frequent item sets have
been listed, this process is repeated. Finding recurring patterns in huge databases is a crucial task.
The likelihood of purchasing subordinate products if a prime item is secured is predicted using two association
rules data mining methods, namely Apriori and Eclat. The Eclat algorithm is quicker than the Apriori method
because of its vertical approach. This algorithm's main goal is to calculate the support of the proposed item set
using set intersection rather than creating subsets that don't already exist in the tree data structure. By altering the
support value, the Eclat approach is run on the dataset. The scanning rate of Eclat is high. It is a clean set because
the span of support and confidence lies between 0 and 1. To test the efficacy of the algorithms, reinforcement
learning techniques like Thompson sampling and upper confidence limit can be used. Classification algorithms
can be added to association rule algorithms to improve them.
In Eclat variations that employ tidset format in the initial looping and diffset in the subsequent looping.
The volume of tidsets is among the key elements determining Eclat's operating time and memory consumption
usage as intersecting tidsets is its primary function. Additional time and storage are required for larger datasets.
Diffset has been demonstrated to significantly outperform traditional Eclat (tidset) in terms of performance and
memory utilization, particularly in large databases. Diffset lacks its strengths over tidsets in sparse databases.
Although the proposed technique performs moderately when it comes to time execution, it displays better patterns
than Eclat-tidset. The algorithm performance would be influenced by the likelihood of itemset repetitions in each
database.

1341
 V. Ravi kumar , Dr. V. Purnachandra Rao

2. Literature Review
In [1], Md. Rezaul Karim et al provided a method for Apache Spark-based MFP mining. Within the Spark
framework, a paradigm is presented for converting a conventional single pipeline to an inter-cluster pipeline. To
generate transaction value databases, two new algorithms named TV and iTV are developed. These algorithms use
prime number-based data processing techniques in both a static and incremental environment. To keep the partly
downward closure property and to decrease the state space and the set of candidate frequent itemsets, two efficient
pruning approaches have been created. Maximal Frequent Pattern with Apache Spark is the name of the suggested
algorithm (MFPAS). For the past few years, the Hadoop MapReduce architecture has been extensively employed
to address some of these issues. SPARK takes a careless approach. Decentralized analytic jobs are managed using
the SOMA platform6, which also supports the big data processing conditions needed for job execution.
In [2], Dingming Wu et al offer a solution based on the FP*-tree, a specially built index structure, and the
corresponding bound-based algorithm. To rapidly determine the top-k maximum diverse frequent itemsets, the
authors merged the maximality and variety requirements. The depth of the class tree is proportional to the width of
each code. Each component of an item's code correlates to a cluster on the path that leads from the base to the
component. The diversity score limit is set at 0 from the beginning. On a 3,040,715 transaction commercial data
set, the proposed techniques are assessed. It uses the Item-Encoding algorithm to determine the diversity value of
each detected maximal frequent item collection. This work proposes a simple approach for mining MDIFs that
incorporates the cutting-edge FPMAX. And the recommended bound-based method performs noticeably better
than the fundamental approach.
In [3], Cagatay Turkay et al introduce Progressive Data Science as a unique knowledge discovery paradigm in
which progress is ingrained in each stage of the data science process. A combined effort from numerous research
communities is necessary to ensure progress in such a novel paradigm. The authors provided a unified perspective
through a reevaluation of the widely used and major KDD pipeline, reviewed the numerous issues raised by
progressiveness, and then discussed some of the promising early efforts from various communities who are
interested in progressive approaches. It is necessary to develop new techniques that can manage progressive
computations because the majority of research so far has concentrated on batch processing settings. These
methods suggest which examples to incorporate into a modeling framework to greatly enhance the precision and
assurance of a learned model.
In [4], Murali Bharatham et al suggested the ARM method for obtaining unit rule matches in the collection of
periodic data collection. The authors identify the non-obscure client raising style in the star bushel analysis.
Additionally, considerations for the goods that would be safeguarded for customers should be made. mining for
associations through social database-driven economic frameworks. Characteristic pillars were placed on various
benches during Flat Partitioning. The entire database is divided into n sections in the first phase. Every wrapped
database is individually stacked into necessary memory before the components of a neighborhood visit are
located. Syndicate every locally visited element, and create a globally welcoming environment. For KDD, the
dynamic inspection point is used to look for relationships within the object sets. On a single processor, the
balanced Apriori calculation's delayed effects work satisfactorily. Working out parallel processors is necessary.
In [5], Bhukya Krishna and Geetanjali Amarawat provided a method for calculating the ongoing thing sets
using the exchanges database's single output in the circle. Because small portions of the database can fit into
primary storage with ease, the approach solves the memory issue for huge databases that cannot fit into primary
memory. The method determines these itemsets' negative border, which is indicated by NBD (S). The set of
candidate itemsets that did not meet minimal support constitutes the negative boundary. The acceptance for the
candidate sets is counted using tidlists. The support for an itemset in a partition is calculated by dividing the
primary key of a tidlists of the itemset by the entire quantity of transactions in the partition. There is a tid list
created for each itemset. The model stores the details of each customer's items in a Customer Bit- vector display
with a sliding window, which is based on the BitVector concept.
In [6], Dongyu Liu et al model multivariate ST data as tensors, and then offer a novel piecewise rank-one
tensor breakdown algorithm that supports automatically dividing the data into homogenous divisions and
identifying the latent features in each split for contrast and visual summarization. A quantitative assessment of
how accurately the derived patterns visually match the actual data is optimized by the algorithm. With the help of
TPFlow, analysts may gradually break the dataset into small subsets along several parameters using a directional
and iterative approach. A subset of the data created during the partitioning process is represented by each node in
the tree. The dataset is more complicated, making it more challenging to directly characterize the original data
using rank-1 factors because the deviations persist even after multiple partitioning rounds. When the vector size
grows, the decomposition process may lose effectiveness. Contextual information is not included in the proposed
method.

1342
 Enhanced Slap Algorithm To Find Top Frequent Itemsets In The Online Retail Store

Table 1: Comparative Mechanisms of the Existing Approaches

S Author Method Merits Demerits
.No
1 Md. Rezaul Apache Spark-based Hadoop MapReduce Many pattern-matching rules
. Karim MFP mining architecture is used.
2 Dingming FP*-tree Includes cutting-edge FPMAX High costs, not scalable, fits
. Wu for smaller datasets.
3 Cagatay Progressive Data provided a unified perspective Only for batch processing.
. Turkay Science using KDD through a reevaluation
4 Murali ARM method FLAT partitioning Not working using parallel
. Bharatham processors as single
processors delays.
5 Bhukya Improved Apriori Splits huge datasets into small Slow computation, scans
. Krishna algorithm clusters and key is issued. dataset many times.
6 Dongyu Liu novel piecewise Homogeneous splits of the data Not scalable, contextual
. rank-one tensor use a directional iterative information is not included.
breakdown algorithm approach so less computation
time.

3. Proposed Methodology
The proposed genetic algorithm is basically divided into two phases: leading and following. The initial
position is known as “Leader”. It updates the position of salp in every iteration as shown in equation (x)
4*t 2
( )
Positionlj  Current _ pos j  2* e L
*(upper _ bou j  lower _ bou j )* rand - (x)

Where,
Positionlj denotes the leader movement in the direction of j th dimension
Current_posj represents the current position of leader
Upper_bou represents the upper boundary of the particle
Lower_bou denotes the lower boundary of the particle
Rand denotes a random variable with a range of 0 to 1
t & L denotes displacement with respect to time
The exploration of domain search space is majorly dependent on the time and displacement. The position
of leader slap is updated with respect to the sources available. The position of the follower slaps is updated using
the equation (x)

1
Follower _ posij  *(curr _ posij  curr _ pos ij1 ) - (x)
2
The proposed research to define a mathematical model integrates the slap algorithm with SVM approach
by tuning the coefficient values of the non-linear kernel functions. The objective function of the algorithms tries
to reduce the error rate along with maximization of accuracy. The pseudo code for computation of objective
function is discussed in below section:
Pseudocode for ESLAP Algorithm:
Input: Online Retail Store Data, ORSD
Output: Accuracy & Top-K utilities

1343
 V. Ravi kumar , Dr. V. Purnachandra Rao

Begin:
1. Initialize slap populations (sp), lower (l) and upper (u) boundaries
2. for i0 to sp-1:
fol_binbinary_conversion(input[i], threshold,sp,len(ORSD))
for j0 to i:
fit[i,j]obj_function(input[i,j], class[i,j], opts)
conrand(0,1)
if(i==0 & con>=0.5):
update the position using equation (x)
elif:
update the follower position using equation (x)
3. Initialize the slap parameters
4. Define model using SVM non-linear kernel
5. acc[0], best_accauc_score(model.fit(X,y))
6. for i 0 to n-1:
acc[i] auc_score(i, model.fit(X,y))
if(acc[i]>best_acc[i]):
best_accacc[i]
4. Results & Discussion
Figure 2 denotes the central tendency metrics over the products to analyze the data based on the quantities
and customer ids. It is observed that most of the transactions has frequent customer patterns.

Figure 2: Descriptive Statistics of Item sets

1344
 Enhanced Slap Algorithm To Find Top Frequent Itemsets In The Online Retail Store

Figure 3: Support & Confidence values for Zinc Product

In Figure 2, the model projects the rules that try to search the frequency using the description “Zinc”. The
equations are presented below:
Lift:
The difference between the actual and expected confidence in an association rule is known as the lift value.
Lift (Zinc => Zinc is a famous mineral) = Confidence (Zinc => Zinc is a famous mineral)
P (Zinc is a famous mineral)

= P (Zinc U Zinc is a famous mineral)

P (Zinc is a famous mineral) P(Zinc)

Confidence:
Confidence is the possibility that when a person purchases item A, he or she is likely to also purchase item B.
Confidence(Zinc-> Zinc is a famous mineral) = Support(Zinc U Zinc is a famous mineral)
Support(Zinc)
Support:
The proportion of groups that contain each of the things stated in an association rule is known as its support.
All of the groupings that were taken into account are used to generate the percentage value.
Support (Zinc) = Number of transactions in which Zinc appears
Total number of transactions

Figure 4: Accuracy Metrics based on Feature Set

1345
 V. Ravi kumar , Dr. V. Purnachandra Rao

Figure 4 denotes the best accuracy obtained with the specified number of iterations. The accuracy of the
model is improved and number of features is reduced.

Figure 5: Fitness Computation based on Number of Iterations

Figure 5 denotes the fitness function generated at each iteration by denoting X-axis with number of iteration
and Y-axis with fitness values.
5. Conclusion
The identification of patterns for finding the frequency of items using the genetic approaches has more impact
than the machine learning. The model initially utilized the concept of encoding units to convert the categorical
data into numerical so that any learning algorithm can train the model easily. Later it visualizes the statistical
representation of the every column to identify the relation in terms of central tendency. When the model is
converted into to numerical the number of attributes are increased and it has to identify the commonness using
the genetic approach known as “SLAP”, which +2.9% better than the existing approaches. It also evaluated by
the other metrics associated with sequence mining like lift, confidence, and support.
References
Karim, M. R., Cochez, M., Beyan, O. D., Ahmed, C. F., & Decker, S. (2018). Mining maximal frequent patterns
in transactional databases and dynamic data streams: A spark-based approach. Information Sciences, 432,
278–300. doi:10.1016/j.ins.2017.11.064
Wu, D., Luo, D., Jensen, C. S., & Huang, J. Z. (2019). Efficiently Mining Maximal Diverse Frequent Itemsets.
Prosody, Phonology and Phonetics, 191–207. doi:10.1007/978-3-030-18579-4_12
Turkay, C., Pezzotti, N., Binnig, C., Strobelt, H., Hammer, B., Keim, D. A., … Rusu, F. (2018). Progressive Data
Science: Potential and Challenges. doi:10.48550/ARXIV.1812.08032
M. Bharatham and Nagalakshmi, "ARM using Apriori with pretend Temporal database portioned Slant in weka
tool," 2018 3rd International Conference on Contemporary Computing and Informatics (IC3I), 2018, pp. 155-
159, doi: 10.1109/IC3I44769.2018.9007278.
Krishna, B., & Amarawat, G. (2018). Data Mining in Frequent Pattern Matching Using Improved Apriori
Algorithm. Emerging Technologies in Data Mining and Information Security, 699–709. doi:10.1007/978-981-
13-1498-8_61
Liu, D., Xu, P., & Ren, L. (2018). TPFlow: Progressive Partition and Multidimensional Pattern Extraction for
Large-Scale Spatio-Temporal Data Analysis. IEEE Transactions on Visualization and Computer Graphics, 1–
1. doi:10.1109/tvcg.2018.2865018

1346
 Enhanced Slap Algorithm To Find Top Frequent Itemsets In The Online Retail Store

Y. Liu, S. Gao, J. Shi, X. Wei and Z. Han, "Sequential-Mining-Based Vulnerable Branches Identification for the
Transmission Network Under Continuous Load Redistribution Attacks," in IEEE Transactions on Smart Grid,
vol. 11, no. 6, pp. 5151-5160, Nov. 2018, doi: 10.1109/TSG.2020.3003340.
Ganesan, M., Shankar, S. High utility fuzzy product mining (HUFPM) using investigation of HUWAS approach. J
Ambient Intell Human Comput 13, 3271–3281 (2019). https://doi.org/10.1007/s12652-021-03231-8.
D. H. Tran, T. T. Nguyen, T. D. Vu, and A. T. Tran, “MINING TOP-K FREQUENT SEQUENTIAL PATTERN
IN ITEM INTERVAL EXTENDED SEQUENCE DATABASE”, JCC, vol. 34, no. 3, p. 249–263, Nov. 2018.
DOI: https://doi.org/10.15625/1813-9663/34/3/13053.
U. Suvarna and Y. Srinivas, “Efficient high-utility itemset mining over variety of databases: a survey,,” in Soft
Computing in Data Analytics, pp. 803–816, Springer, Berlin, Germany, 2019. DOI: 10.1007/978-981-13-
0514-6_76.
Lin, J.C.-W., Djenouri, Y., Srivastava, G., Li, Y., Yu, P.S.: Scalable mining of high-utility sequential patterns
with three-tier MapReduce model. ACM Trans. Knowl. Discov. Data 16(3), 60:1–60:26 (2019) DOI:
https://doi.org/10.1145/3487046.
Nunez-del-Prado, M.; Maehara-Aliaga, Y.; Salas, J.; Alatrista-Salas, H.; Megías, D. A Graph-Based Differentially
Private Algorithm for Mining Frequent Sequential Patterns. Appl. Sci. 2019, 12, 2131.
https://doi.org/10.3390/app12042131.
Dahiya, Vandna, and Sandeep Dalal. "A Systematic Literature Review of Utility Itemset Mining Algorithms for
Large Datasets." REVISTA GEINTEC-GESTAO INOVACAO E TECNOLOGIAS 11.3 (2019): 565-582.
V. S. Tseng, B. E. Shie, C. W. Wu, and P. S. Yu, ‘‘Efficient Algorithms For Mining High Utility Itemsets From
Transactional Databases,’’ IEEE Transactions Knowledge and Data Engineering, volume 25, number 8, pp.
1772–1786, 2013, https://ieeexplore.ieee.org/document/6171188.
Sathyavani, D., Sharmila, D. RETRACTED ARTICLE: An improved memory adaptive up-growth to mine high
utility itemsets from large transaction databases. J Ambient Intell Human Comput 12, 3841–3850 (2019).
https://doi.org/10.1007/s12652-020-01706-8.
R. Agarwal, A. Gautam, P. Dixit and A. Rana, "An Approach to Mine Frequent Item Sets Considering Negative
Item Values," 2020 8th International Conference on Reliability, Infocom Technologies and Optimization
(Trends and Future Directions) (ICRITO), 2020, pp. 208-211, doi: 10.1109/ICRITO48877.2020.9197870.
Singh, K., Kumar, R. & Biswas, B. High average-utility itemsets mining: a survey. Appl Intell 52, 3901–3938
(2019). https://doi.org/10.1007/s10489-021-02611-z.
Chunkai Zhang, Zilin Du, Wensheng Gan, Philip S. Yu, TKUS: Mining top-k high utility sequential patterns,
Information Sciences, Volume 570, 2019, Pages 342-359, ISSN 0020-0255,
https://doi.org/10.1016/j.ins.2021.04.035.

1347