A model for profit pattern mining based on genetic algorithm

IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
A MODEL FOR PROFIT PATTERN MINING BASED ON GENETIC
ALGORITHM
Vivek Badhe1
, R.S. Thakur2
, G.S. Thakur3
1
Research Scholar, Maulana Azad National Institute of Technology, Bhopal, India
2
Associate Professor, Maulana Azad National Institute of Technology, Bhopal, India
3
Assistant Professor, Maulana Azad National Institute of Technology, Bhopal, India
Abstract
Mining profit oriented patterns is a novel technique of association rule mining in data mining, which basically focuses on
important issues related with business. As it is well known that every business aims to generate the profit and find the ways to
improve the same. In earlier days association rule mining was used for market basket analysis and targeted only some of the
business and commercial aspects. Afterwards the researchers started to aim the most prominent element of any business i.e.
Profit, and determined the innovative way to generate the association rules based on profit. Profit oriented patterns mining
approach combines the statistic based pattern mining with value-based decision making to generate those patterns with the
maximum profit and some ways to generate recommenders for future strategy. To achieve the desired goal the traditional
association rule mining alone is not effectual, so we combine the strength of genetic algorithm with association rule mining to
enhance its capability. The study shows that Genetic Algorithm improves the effectiveness and efficiency of association rule
mining outcome, since genetic algorithms are competent to handle the problems related with the uncertainty, multi-
dimensional, non-differential, non-continuous, and non-parametrical, non-linearity constraint and multi-objective optimization
problems. In this paper we apply the concept of profit pattern mining with genetic algorithm to generate profit oriented pattern
which help out in future business expansion and fulfill the business objective.
Keywords: Data Mining, Association Rule Mining, Profit Pattern Mining, Genetic Algorithm
-------------------------------------------------------------------------
***------------------------------------------------------------------------
1. INTRODUCTION
Data mining [1] refers to the discovery of new information
in terms of patterns or rules from vast amount of data. It the
technique basically used for extracting the hidden predictive
information from large database repository. It is a modern
and powerful methodology with immense potential to
analyze significant information from the huge databases.
Discovery of unseen pattern is an essential database-mining
task. While being a vital tool for several practitioners, data
mining is also an attractive research area that raises many
challenging problem [7].
Data Mining is the methodology to finding significant, new
correlation pattern and trends by sifting through huge
amount of data stored in repository, using patterns
discovery technique as well as statistical and mathematical
techniques. The two primary aims of data mining [2] are
Prediction and description. Prediction makes use of existing
variables in the database in order to predict unknown or
future values of interest, and description focuses on finding
patterns describing data and the subsequent presentation for
user interpretation. The relative emphasis of the both
prediction and description differ with respect to the
underlying application and the technique. There are several
data mining techniques such as classification, clustering,
outlier analysis and association rule mining fulfilling these
objectives.
However data mining sometimes viewed as s a multi-
objective task the Genetic Algorithm [3] with rule mining
achieve the target. Since data mining may also be viewed
as the process of turning the data into information, the
information into action, and action into value or profit, and
profit pattern mining is the way to achieve this by using
profit as measure of interest.
2. PRELIMINARIES
2.1 Association Rule Mining
Extraction of Association rules is one of the most important
techniques which were given by R. Agrawal et. al. in 1993
[4]. It provides the information like "if-then" statements.
These rules are invoked from the dataset. It generates from
calculation of the support and confidence of each rule that
can show the frequency of occurrence of a given rule.
Association Analysis is the method of discovering unseen
pattern or correlation condition that occurs frequently
together in a given dataset but not visible due to large
volume of data. Association Rule mining techniques looks
for interesting associations and correlations among data set.
An association rule is a rule, which entails probabilistic
relationship, with the form X ⇒ Y between sets of database
attributes, where X and
Y are sets of items, and X ∩ Y = . Given the set ofϕ
transactions T, we are interested in generating all rules that
satisfy certain constraints. These constrains are support and
confidence. The support of the rule is the fraction of the
transactions in T that satisfy the union of items in X and Y.
The probability, measured as the fraction of the transactions
containing X also containing Y, is called the confidence of
the rule. Confidence is a measure of the rule's strength,
_______________________________________________________________________________________________________
Volume: 05 Issue: 08 | Aug-2016, Available @ http://ijret.esatjournals.org 43

support corresponds to statistical importance. With the help
of these constraints, rules are computed from the data and,
association rules are calculated with help of probability.
Mining frequent itemsets [7] is a fundamental and vital task
in many data mining applications such as the discovery of
association rules, strong rules, correlations, multi-
dimensional patterns, and many other important discovery
tasks. The most important and principal algorithm that was
given to generate association rules was apriori [5].
2.2 Profit Pattern Mining
The goal of Profit Pattern Mining (PPM) [8] is to develop a
model which generates profitable rules, recommender rules
that recommends target items and promotion strategy for
future customer. It is a novel way of association Rule
Mining which aims to find out those patterns which
provides maximum profit. As the major obstacle in the
association Rule mining application is the gap between the
statistical based patterns extraction and valued based
decision making, Profit Pattern mining reduces this gap.
Fig 1: Hierarchy of Profit Pattern Mining
The Hierarchy of PPM shown in figure 1, In Profit Pattern
Mining a set of preceding transaction and preferred target
item is given and build a model for recommending target
items and marketing strategies to new customers, with the
aim of maximizing the business profit
2.3 Genetic Algorithm
The Theory Genetic Algorithm (GA) was given by John
Holland in 1970. It incorporates Darwinian evolutionary
theory survival of the fittest. Genetic algorithm [6] is a type
of searching algorithm. It looks for an entire solution space
for an optimal result to a problem. The main feature of the
genetic algorithm is how the searching is made. The
algorithm creates a “population” of possible solutions to the
problem and lets them “evolve” over multiple generations to
find better and better solutions. The algorithm operates
through a simple cycle Population creation of strings,
Evaluation of each string, Best string selection, Genetic
manipulation to create a new population of strings. Figure
below shows the interconnection of these four stages. Each
cycle (show in figure 2) produces a new generation of
possible solutions (individuals) for a given problem.
Fig 2: Genetic Algorithm Cycle
The manipulation process enables the genetic operators to
produce a new population of individual the offspring, by
manipulation the genetic information processed by the pairs
chosen to reproduce. The information stored in strings
(chromosomes) that describes the individuals. Genetic
operators are used. The offspring generated by this process
take the place of the older population and the cycle is
repeated until a desired level of fitness is attained, or a
determined number of cycles are reached.
The algorithm begins by creating a random initial
population. The algorithm then produces a sequence of next
population, or generations. At each step, the algorithm uses
the individuals in the current generation to create the next
generation. For creating the subsequently generation, the
algorithm executes the following iterative steps: Scores each
member of the present population by computing its fitness
value. Scale the raw fitness scores to convert them into a
more usable range of values. Selects parents based on their
fitness. Produce children from the parents. Children are
produced either by making random changes to a single
parent – mutation – or by combing the vector entries of a
pair of parents – Crossover. Replace the current population
with the children to form the next generation. The
algorithm stops when one of the five stopping criteria ie
generations, time limit, fitness limit, stall generations or
stall time limit is met and finally the ultimate beauty of
Genetic Algorithm [16] is adaptability itself.
3. ASSOCIATION RULE MINING WITH
GENETIC ALGORITHM
The utility of the genetic algorithm in the perspective of
data mining is that Genetic Algorithms are robust,
amendable and they approach uniformly to large number of
different classes of data mining problems [3]. If the solution
for given problems exists, the Genetic Algorithms with
proper coding, operators and fitness function will find it.
This is an obvious advantage over traditional methods and
models that can only be used in specific cases. Such
generality is desirable in Data Mining where the search
space is complex noise.
Genetic Algorithm is used with Data mining in one of the
three different ways: GA in Pre-Mining: Although Genetic
Algorithm is rarely applied in pre-mining ie. Before mining
process, still for some non linear constraints the GA is used
in a pure traditional way. GA in In-Mining: Genetic
Algorithm is applied during mining process in two ways:
_______________________________________________________________________________________________________

Traditional Genetic Algorithm: During the mining process
Genetic Algorithm is applied between the mining methods
ie. Association rule mining, clustering or outlier analysis
etc. to optimize the results. The traditional GA is rarely
used in-mining process. Modified Genetic Algorithm:
During the mining process Genetic Algorithm itself is
modified ie. The steps of Genetic Algorithm are altered
accordingly and this is commonly used approach in in-
mining process. GA in Post mining: In post mining, first
the traditional mining is done and then Genetic Algorithm
is applied to optimize the results. In general Genetic
Algorithm is applied in a traditional way but sometimes the
encoding method is modified as per the need. The use of
Genetic Algorithm in Post mining is most commonly done.
Association Rule Mining uses primarily Apriori method for
generating rules by using statistical measures, but there is
no surety whether these rules formed are best at hand. A
proposition was given by Manish Saggar et al. [9]
suggesting that the induction of Genetic Algorithm will
facilitate the optimization of rules being generated by
Apriori while also taking into consideration the negative
incidence. Now, optimizing the rules is one thing but
generating optimized rules is another. The two metrics
syntactic and transactional superiority proposed in 2008
[10] serves as the plummet in ascertaining the existence of
interesting rules form a resultant set of rules. These two
metrics are responsible for maintaining the quality of the
optimized interesting rules generated in spite of the
dominance of the regular rules of Apriori. Subsequently, the
algorithm takes a bit longer to generate optimized rules and
allows a better representation of results to the users.
The Apriori algorithm and is extensions like Partition,
Pincer-Search, Incremental etc, uses properties that allow
only those objects (attributes) to be candidates that acquire a
certain threshold. Hence, the objects that do not fit the
criteria are eliminated from the algorithm, and there is no
sure way to keep these objects with negative occurrences.
To find all the possible rules being generated form a given
dataset GA is applied [11]. If Genetic Algorithm (GA) is
used, a predictive analysis could be done for the generated
rules that contain negative objects. Also these rules have
more than one object in consequent part as compared to the
regular ones. The predictive rule discovery using GA proves
to be advantageous as it performs a global search and is
based on the greedy approach thus have less complexity as
compared to other algorithms. The traditional association
rule algorithms use support and confidence as threshold
value to generate interesting rules. The genetic algorithm-
based strategy designed by Xiaowei Yan et al. [12]
identifies association rules without considering the actual
support threshold. Their approach employs an elaborated
encoding method where the relative confidence threshold is
used as the fitness function for the algorithm. As the
proposed model did not require minimum support
threshold, the GA performs a global search and a system
automation in enforced.
The GA algorithm has less complexity as compared to other
algorithms because it follows the greedy approach.
Soumadip Ghosh [13] using this property mines frequent
itemsets using GA. The frequent itemsets discovered by GA
is done by performing a global search on the dataset. In
2012, K Indira and S Kanmani [15] proposed a technique
which was able to analyze the performance of Genetic
Algorithm when used for Mining ARs. The works
considered for analysis were over a period of seven years for
mining ARs using GA. The performance analysis of GA
was done effectively by making a small yet beneficial
amount of modifications to GA operators and monitor how
the parameter functions in some particular interim.
4. RELATED WORK
In 2002 Ke Wang, Senqiang Zhou, and Jiawei Han
presented a concept of profit mining [8], this approach to
reduce the gap between the statistic-based pattern mining
and the value-based decision making. They obtained a set of
past transactions and pre-selected target items, and intended
to construct a model for recommending target items and
promotion strategies to new customers, with the goal of
maximizing the net profit. They identified some issues in
profit mining and proposed solutions. They evaluate the
effectiveness of this approach using data sets of a wide
variety of characteristics. The key to profit mining is to
suggest “right” items and “right” cost. If the price is too
high, the customer will go away without generating any
profit; if the price is too low or if the item is not profitable,
the profit will not be maximized. The approach is to exploit
data mining to discover the patterns for right items and
right cost. The major issues in this context are Profit
oriented patterns, shopping on unavailability, huge search
space, optimality and interpretability of recommender.
Another efficient approach based on weight factor and
utility for proficient mining of significant association rules
proposed by Sandhu, P.S. et. al. in 2010 [14]. Initially, the
proposed approach makes use of the conventional Apriori
algorithm to generate a set of association rules from a
repository. The proposed approach exploits the anti-
monotone property of the Apriori algorithm, which states
that for a k-itemset to be frequent all (k-1) subsets of this
itemset also have to be frequent. Subsequently, the set of
association rules mined are subjected to weight age (W-
gain) and utility (U-gain) constraints, and for every
association rule mined, a combined Utility Weighted Score
(UW-Score) is computed. Ultimately, they determined a
subset of valuable association rules based on the UW-Score
computed. The experimental results show the effectiveness
of the proposed approach in generating high utility
association rules that can be profitably applied for business
growth
5. PROPOSED WORK
The proposed framework and design shown in figure 3 for
mining the profit patterns using Genetic Algorithm covers
the following tasks and they are Data Preprocessing,
Implementing ARM Algorithm on processed data and
optimizing the Rules using GA. Initially we have perform
the conventional association rule mining algorithm
implemented in C++ language to generate rules from the
preprocessed database than we optimized those generated
rules by Genetic Algorithm implemented in GA Solver of
optimization toolbox called optimtool of MATLAB®.
_______________________________________________________________________________________________________

Fig 3: Block Diagram of Methodology
To design the fitness faction for the genetic algorithm we
use measure called profit and define the notion of profit
because the profit in any business is the key element and the
notion of profit may vary depending upon the type of
business but in general the notion of profit could be
categorized as under:
Value profit: Value profit is simply the difference of selling
price and the cost price of any product. It is also called the
margin of profit.
Percentage of profit: It is the percentage of margin of
profit with respect to the cost price of any product.
Quantitative profit: It is the profit based on the number of
items sold, and sometimes it is known as weighted factor.
For each rule
Item1  Item2
The fitness function designed as
Where w1 and w2 are the user defined weighted factors,
and C and I namely Completeness and Interestingness. The
value of w1 and w2 are calculated as:
Completeness and Interestingness are defined & calculated
as:
Completeness (C): Those rules are considered as complete
rules where:
Item1Item2
Item1 is having lower percentage of profit & Item2 having
higher percentage of profit.
Interestingness (I): Those rules are considered as rules of
interest where
Item1Item2
Item1 is having lower value profit & Item2 having higher
value profit.
Where TP, FP, FN are defined as
(For sample rule Item1 Item2)
True Positive (TP):
No. of rules satisfying both Item1& Item2
False Positive (FP) :
No. of rules not satisfying Item1 but satisfying Item2.
False Negative (FN):
No. of rules satisfying Item1 but not satisfying Item2.
Pseudo code of proposed methodology
1. Start
2. Preprocess the Row Dataset |RD|
3. Transform the |RD| to Relevant Transaction Dataset |D|
4. Load the Sample Transactions |S| from Dataset |D|
5. Apply Apriori Algorithm to |S| for Rule Generation
with defined parameter Support and Confidence.
6. Store the output of Apriori to rule set |R|
7. Apply the GA Cycle on |R|
i. Selection - Tournament
ii. Crossover – Single Point
iii. Mutation - uniform
iv. Check fitness – Defined Fitness Function FF
v. Check termination Condition Stall 100
8. Store the outcome of GA as final result to |F|, which
contains the optimized (profitable) rules.
9. Mapped the |F| Rule with desire format
10. Stop
6. IMPLEMENTATION AND RESULT
The dataset that we gathered from departmental store
contains retail data of Fast Moving Consumer Goods
(FMCG) during sales of Second-Third quarter of year 2003.
The dataset contains 9 attributes and 16293 records. The
records hold repeated yet different transactions for a
number of products. The 9 attributes of the dataset are
BillNo, ProductCode, ProductName, Packaging,
ProductCategory, Quantity, PurchasedPrice, SellingPrice,
BillDate. All 16293 records have a number of items
purchased on a distinct bill.
RawData: The above dataset is then pre-processed in order
to be more relevant for the mining task. This is done by first
finding the number of distinct BillNo, ProductCode,
ProductName and arranging all column-wise in ascending
order. On processing the dataset the findings are listed in
Table 1.
_______________________________________________________________________________________________________
Row
Data
Data
Pre-processing
Data
Transformation
Classical
ARM
Genetic
Algorithm
with Profit
Fitness
Function
Profit
Pattern
s

Table 1: Dataset Description
S.No. Category Count
1 Product List 16293
2 BillNo 1703
3 ProdName 1743
4 ProdCodes 2249
Firstly the database is taken which is shown below and
converted into Flat file text tab delimited format as show in
figure 4.
Fig 4: Snapshot of Row Data
Applying Apriori algorithm on the processed data and
generating the following rules shown in figure 5.
Fig 5: Snapshot of Apriori Generated Rule
Now using Genetic Algorithm toolbox shown in figure 6 in
MATLAB Version 7.6.0.324 (R2008a) above rules are
optimized to produce the desired profit oriented rules,
table2 and table3 shows the generic and profit rules
respectively.
_______________________________________________________________________________________________________

Fig 6: Snapshot of MATLAB Toolbox
Table 2: Generic Rule with Fitness value
Rules C I W1 W2 Fitness
AAG111 <-- AAG005 1.00 0.50 0.13 0.11 0.8
AAG292 <-- AAG235 1.00 0.60 0.23 0.07 0.9
Table 3: Profit Oriented Rule with Fitness value
Rules C I W1 W2 Fitness
AAG111 <-- AAG005 1.00 0.50 0.13 0.11 0.8
AAG220 <-- BAG049 1.00 0.25 0.21 0.52 0.5
AAG105 <-- AAG005 1.00 0.50 0.31 0.34 0.7
AAG295 <-- AAG292 0.67 0.40 0.06 0.06 0.5
AAG295 <-- AAG235 0.50 0.20 0.13 0.04 0.4
AAG296 <-- AAG292 0.67 0.50 0.13 0.13 0.6
AAG296 <-- AAG235 0.33 0.20 0.30 0.09 0.3
BAG186 <-- BAG049 0.13 0.25 0.10 0.05 0.2
AAG154 <-- BAG049 0.14 0.25 0.06 0.15 0.2
AAG198 <-- BAG049 0.25 0.25 0.09 0.09 0.3
BAG049 <-- AAG198 0.50 0.13 0.11 0.11 0.3
BAG186 <-- AAG010 0.10 1.00 0.10 0.08 0.5
AAG235 <-- AAG292 1.00 0.20 0.04 0.14 0.4
AAG292 <-- AAG235 1.00 0.60 0.23 0.07 0.9
AAG154 <-- BAG186 0.29 0.33 0.06 0.30 0.3
BAG186 <-- AAG154 0.50 0.71 0.17 0.03 0.5
AAG198 <-- BAG186 0.50 0.33 0.09 0.19 0.4
BAG186 <-- AAG198 0.29 0.29 0.11 0.05 0.3
AAG198 <-- AAG154 0.14 0.25 0.16 0.06 0.2
AAG154 <-- AAG198 0.57 0.50 0.06 0.16 0.5
AAG295 <-- AAG292 AAG235 0.33 0.50 0.41 0.03 0.3
AAG296 <-- AAG292 AAG235 0.33 0.50 0.91 0.05 0.3
_______________________________________________________________________________________________________

BAG049 <-- AAG154 AAG198 0.50 0.50 0.66 0.04 0.5
AAG010 <-- BAG186 AAG154 0.67 0.67 0.63 0.03 0.7
AAG010 <-- BAG186 AAG198 0.67 0.67 0.51 0.04 0.7
AAG198 <-- BAG186 AAG154 0.25 0.33 0.58 0.05 0.3
AAG154 <-- BAG186 AAG198 0.20 0.33 0.30 0.10 0.2
BAG186 <-- AAG154 AAG198 0.17 0.33 0.67 0.02 0.2
AAG010 <-- BAG186 AAG154 AAG198 0.33 1.00 0.40 0.03 0.4
7. CONCLUSION
Association Rule Mining for profit pattern combine the
statistic based pattern extraction with value-based decision
making to achieve the commercial goals. In our proposed
approach we have propose a model in which we apply
classical association rule mining followed by genetic
Algorithm techniques. Genetic Algorithm not only
improves the mining process but also provide the optimized
rules. Although a several researches has been carried out in
association rule mining but still it requires more attention
for defining the notion of profit which would help in
improving business strategy and better decision making
ACKNOWLEDGMENT
This work is supported by research project under Fast Track
Scheme for Young Scientist from DST, New Delhi, India.
Scheme 2011-12, No. SR/FTP/ETA-121/ 2011 (SERB),
dated 18/12/2012.
REFERENCES
Books
[1] J. Han and M. Kamber, “Data Mining: Concepts and
techniques”, Morgan Kaufmann Publishers, Elsevier
India, 2001.
[2] A. K. Pujari, Data Mining Techniques, University
Press 2001.
Journals
[3] Satchidananda Dehuri, Ashish Ghosh, R Mall, Parallel
Multi-objective Genetic Algorithm for Classification
Rule Mining, IETE Journal of Research,vol 53,No .5,
PP475-483
[4] R Agrawal, T.Imielinski, and A.Swami, 1993.
“Mining association rules between sets of items in
large databases”, in proceedings of the ACM SIGMOD
Int'l Conf. on Management of data, pp. 207-216.
[5] R.Agrawal and R.Shrikanth.“Fast Algorithm for
Mining Association Rules.” In Proceeding of VLBD
Conference, Santigo, Chile-1994,pp 487-494
[6] Melanie Mitchell, An Introduction to Genetic
Algorithms, PHI, 1996
[7] A. Tiwari, R.K. Gupta and D.P. Agrawal “A survey on
Frequent Pattern Mining : Current Status and
Challenging issues” Information Technology Journal
9(7) 1278-1293, 2010.
[8] Ke Wang, Senqiang Zhou, and Jiawei Han, Profit
Mining: From Patterns to Actions, C.S. Jensen et al.
(Eds.): EDBT 2002, LNCS 2287, pp. 70–87,
2002.Springer-VerlagBerlin.
[9] Manish Saggar, Ashish Kumar Agarwal and
Abhimunya Lad, “Optimization of Association Rule
Mining using Improved Genetic Algorithms”IEEE
2004
[10] Peter P. Wakabi-Waiswa , Dr. Venansius
Baryamureeba, “Extraction of Interesting Association
Rules Using Genetic Algorithms”, Advances in
Systems Modelling and ICT Applications, pp. 101-
110. G
[11] Anandhavalli M, Suraj Kumar Sudhanshu, Ayush
Kumar and Ghose M.K., “Optimized association rule
mining using genetic algorithm”, Advances in
Information Mining, ISSN: 0975–3265, Volume 1,
Issue 2, 2009, pp-01-04.
[12] Xiaowei Yan, Chengqi Zhang, Shichao Zhang,
“Genetic algorithm-based strategy for identifying
association rules without specifying actual minimum
support”, Expert Systems with Applications 36 (2009)
3066–3076
[13] Soumadip Ghosh, Sushanta Biswas, Debasree Sarkar,
Partha Pratim Sarkar“Mining Frequent Itemsets Using
Genetic Algorithm”, International Journal of Artificial
Intelligence & Applications (IJAIA), Vol.1, No.4,
October 2010
[14] Sandhu, P.S.; Dhaliwal, D.S.; Panda, S.N.; Bisht, A.,
“An Improvement in Apriori Algorithm Using Profit
and Quantity” ICCNT Year: 2010, IEEE conference
publication.
[15] Indira K, Kanmani S, Performance Analysis of Genetic
Algorithm for Mining Association Rules, International
Journal of Computer Science Issues, Vol. 9, Issue 2,
No 1, March 2012 ISSN (Online): 1694-0814.
[16] Mehmed Çelebi, A new approach for the genetic
algorithm, Journal of Statistical Computation and
Simulation, Taylor & Francis group, volume 79, issue
3 March 2009,, PP 275-279
_______________________________________________________________________________________________________

A model for profit pattern mining based on genetic algorithm

More Related Content

What's hot (19)

Similar to A model for profit pattern mining based on genetic algorithm (20)

More from eSAT Journals (20)

Recently uploaded (20)

A model for profit pattern mining based on genetic algorithm