COMPACT WEIGHTED CLASS ASSOCIATION RULE MINING USING INFORMATION GAIN

International Journal of Data Mining & Knowledge Management Process (IJDKP) Vol.1, No.6, November 2011
DOI : 10.5121/ijdkp.2011.1601 1
COMPACT WEIGHTED CLASS ASSOCIATION RULE
MINING USING INFORMATION GAIN
S.P.Syed Ibrahim1
and K.R.Chandran2
1
Assistant Professor, Department of Computer Science and Engineering,
PSG College of Technology, Coimbatore, India
sps_phd@yahoo.co.in
2
Professor of Information Technology and Head, Department of Computer and
Information Sciences, PSG College of Technology, Coimbatore, India
chandran_k_r@yahoo.co.in
ABSTRACT
Weighted association rule mining reflects semantic significance of item by considering its weight.
Classification constructs the classifier and predicts the new data instance. This paper proposes compact
weighted class association rule mining method, which applies weighted association rule mining in the
classification and constructs an efficient weighted associative classifier. This proposed associative
classification algorithm chooses one non class informative attribute from dataset and all the weighted class
association rules are generated based on that attribute. The weight of the item is considered as one of the
parameter in generating the weighted class association rules. This proposed algorithm calculates the
weight using the HITS model. Experimental results show that the proposed system generates less number of
high quality rules which improves the classification accuracy.
KEYWORDS
Weighted Association Rule Mining, Classification, Associative Classification.
1. INTRODUCTION
Data mining principally deals with extracting knowledge from data, In the world where data is all
around us, the need of the hour is to extract knowledge or interesting information, which is
hidden in the available data. Association rule mining is concerned with extracting a set of highly
correlated features shared among a large number of records in a given database. It uses
unsupervised learning where no class attribute is involved in finding the association rule.
Although classical association rule mining algorithm reflects the statistical relationship between
items, it does not reflect the semantic significance of the items [10]. To meet the user objective
and business value, various weighted association rule mining methods [17] [15] [16] were
proposed based on the weightage to items.
On the other hand, classification uses supervised learning where class attribute is involved in the
construction of the classifier. Both, weighted association rule mining and classification are
significant and efficient data mining techniques. So integration of these two data mining
techniques may provide efficient associative classifier [18].
In [18] syed et al., have proposed weighted associative classification based on class based
association (CBA) [13]. Weighted associative classification (WAC) algorithm pre-assigns the
weight for each item randomly. WAC algorithm generates huge number of rules, so it suffers
with high computation cost.
To improve the performance, recently the authors proposed compact weighted associative
classification (CWAC) [19]. The CWAC algorithm is completely varies from WAC. In WAC,
Apriori association rule mining algorithm is directly applied to find the class association rules,

2
whereas CWAC algorithm chooses one non class attribute randomly from dataset and all the
items are generated only based on that attribute. In this way CWAC algorithm reduces the
number of itemset generation and it assignes the weight for each item using HITS model [11].
CWAC algorithm calculates the weighted support and weighted confidence for each item and
determines whether the item was frequent or not. This paper aiming to improve the performance
of the CWAC algorithm and evaluates the effect of weighted associative classification method in
the various bench mark datasets.
The rest of the paper is organized as follows: Section 2 gives an insight about the past work in
this field and section 3 explains the attribute selection strategy, rule generation and rule
evaluation strategies. Section 4 presents the experimental results and observations followed by
the conclusion.
2. RELATED WORK
2.1 Association Rule Mining
Association Rule Mining (ARM) [1], has become one of the important data mining tasks. ARM is
an unsupervised data mining technique, which works on variable length data, and it produces
clear and understandable rules. The basic task of association rule mining is to determine the
correlation between items belonging to a transactional database. In general, every association rule
must satisfy two user specified constraints, one is support and the other is confidence. The
support of a rule X Y (X and Y are items) is defined as the fraction of transactions that contain
X and Y, while the confidence is defined as the ratio support(X and Y)/support(X). So, the target
is to find all association rules that satisfy user specified minimum support and confidence values.
2.2. Weighted Association Rule Mining
Classical ARM framework assumes that all items have the same significance or importance i.e.
their weight within a transaction or record is the same (weight=1 per item) which is not always
the case. In the supermarket context, some items like jewellery, designer clothes, etc., are of
much significant in terms of revenue or profit by the store. Hence weight can be used as a
parameter to generate association rule mining called as weighted association rule mining [17, 16].
The weighted association rules are generated based on user specified minimum weighted support
and minimum weighted confidence thresholds. The use of weighted support and weighted
confidence leads to useful mechanisms to prioritize the rule according to their importance, instead
of their support and confidence alone.
2.3. Classification
Construct the classifier based on training dataset and predicts the class object for new dataset.
Classification uses supervised learning where class attribute is involved in constructing a
classifier.
2.4. Associative Classification
Associative classification was first introduced by Liu et al[13] which focus on integrating two
known data mining tasks, association rule discovery and classification. The integration done is
focused on a special subset of association rules whose right hand side is restricted to the class
attribute; for example, consider a rule R: X Y, Y must be a class label. Associative
classification generally involves two stages. In the first stage, it adopts the association rule
generation methods like Apriori candidate generation [2], or FP growth [8] algorithms to generate
class association rules.

3
For example CBA [13] method employs Apriori candidate generation [2] and other associative
methods such as CPAR [24], CMAR [12] while Lazy associative classification [3, 4] methods
adopts FP growth algorithm [8] for rule generation. The rule generation step generates huge
number of rules. Experimental results reported by Baralis et al., [4] has shown that the CBA
method which follows Apriori association rule mining algorithm generates more than 80,000
rules for some datasets that leads to memory exceptions and other severe problems, such as
overfitting etc.,
If all the generated rules are used in the classifier then the accuracy of the classifier would be high
but the process of classification will be slow and time-consuming. So in the next stage, generated
rules are ranked based on several parameters and interestingness measures such as confidence,
support, lexicographical order of items etc. Then only the high-ranking rules are chosen to build a
classifier and the rest are pruned.
Evolutionary based associative classification method [22] was proposed recently. This approach
takes subset of rules randomly to construct the classifier. Richness of the ruleset was improved
over the generation.
Syed et al., [21] proposed Lazy learning associative classification where it classifies the new data
sample without constructing the classifier but this lazy approach results in high CPU utilization
time and cost.
Chen et al., [7] and Zhang et al., [25] proposed a new approach based on information gain where
more informative attribute are chosen for rule generation. An informative attribute centred rule
generation produces a compact ruleset.
Syed et al.,[20] proposed compact weighted associative classification based on information gain,
where class association rule generation algorithm chooses information gain non class attribute
from dataset and all the items are generated only based on that attribute. Thus this algorithm
reduces number of itemset generation. Finally the algorithm calculates the weighted support and
weighted confidence for each item and determines whether the item is frequent or not.
In [19] the author’s proposed genetic network based associative classification method, which
generates sufficient number of rules to construct the classifier. Here information gain attribute is
used to construct the compact genetic network.
2.5. Weighted associative classification
Syed et al., [18] have proposed weighted associative classification (WAC), which integrates
weighted association rule mining and classification to construct the efficient weighted associative
classifier. Weighted associative classifier extracts special subset of association rules called
weighted class association rules (WCARs). Weighted association rule mining uses weight as one
parameter but here weights for each item item are assigned randomly but it is very difficult to
assign weights to each item.
2.6. HITS Model
Kleinberg [11] used HITS algorithm in bipartite graph and weights are derived from the internal
structure of the database. Sun et al., [17] uses this HITS model to derive weights for each item in
the dataset and derived the weighted association rules. This proposed method uses HITS model
to derive the weight for each item. Then these weights are used to compute the Weighted Class
Association Rules.

4
3. PROPOSED SYSTEM
3.1. Problem Definition
Let database D is a set of instances where each instance is represented by < a1, a2 …am , C>,
where a1, a2 …am, are attributes and C are class value each has weights. A common rule is
defined as x → c, where x is a set of non class attributes and c is class label. The quality
measurement factor of a rule is weighted support and weighted confidence. Rule items that satisfy
minimum weighted support and weighted confidence are called frequent weighted rules, while
the rest are called infrequent weighted rules. Here the task is to generate the Weighted Class
Association Rules (WCARs) that satisfies both minimum weighted support and minimum
weighted confidence constraints. Then these WCARs are used to construct a classifier based on
Confidence, Support, and size-of-the rule Antecedent.
Figure 1. The proposed System
3.2 Attribute Selection based on Information Gain
Generally Apriori based rule generation algorithm generates 2k rules for the dataset with K items.
To reduce the rule generation Information gain attribute is used. Information gain is a measure
that is used in information theory to quantify the ‘information content’ of messages [9]. In ID3
decision tree algorithm [14] information gain is used to choose the best split attribute.
Information gain measure is used to identify the best split attribute in decision tree classifier. In
this paper, it could be used to generate the class association rules. In the process of generating the
class association rules, instead of considering all the attributes, information gain measure will be
Choose the Information Gain Attribute
Generate class association rules based on information gain attribute
Weighted Class association rule generation based on weighted support and weighted
confidence
Rank Ranking and Rule Pruning
Construct the Classifier
Predict the Class
Predict the new dataset

5
used to select the best splitting attribute. In this way, the proposed work generates the limited
number of good quality rules.
Suppose an attribute A has n distinct values that partition the dataset D into subsets T1, T2, ..., Tn.
For a dataset, freq(Ck, D) / |D| represents the probability than an tuple in D belongs to class Ck.
Then info(D) is defined as follows to measure the average amount of information needed to
identify the class of a transaction in D:
Info(D)=
||
),((
log
||
),(
2
1 D
DCfreq
X
D
DCfreq k
g
k
k
∑=
− (1)
Where |D| is the number of transactions in database D and g is the number of classes. After the
dataset D is partitioned into n values of attribute A, the expected information requirement could
be defined as:
InfoA (D) ∑= X
D
Di
||
||
info(Di) (2)
The information gained by partitioning D according to attribute A is defined as:
Gain (A) = info (D) – info A (D) (3)
The best split attribute is the one that maximized the information gain in the data set D. These
best attributes is used to generate the subset.
3.3 Subset Generation
After identifying the information gained attribute, the subsets are generated only based on
information gain attribute. Let us consider Table I, contains 14 transaction, and 2 class values.
Among the four attributes, attribute ‘CD4 Cell Count’ is selected as the informative attribute as it
has the maximum IG value.
TABLE 1 Sample Dataset

6
So for the above dataset the following subset can be created for the scanning of the first
transaction. Generated rules using information gain based associative classification are
{CD4 Cell Count >500} AIDS =No}
{CD4 Cell Count > 500, Sweating at Night = High AIDS = No}
{CD4 Cell Count > 500, Sweating at Night = High, Tuberculosis (TB) = No
AIDS= No}
{CD4 Cell Count > 500, Sweating at Night = High, Tuberculosis (TB) = No,
Temperature == Normal AIDS = No}
The other itemsets that are commonly generated by the Apriori based association rule mining
procedure are eliminated. The excluded rules are
{Sweating at Night == High AIDS = No }
{Tuberculosis == No AIDS = No }
{Temperature == Normal AIDS = No }
{Sweating at Night == High, Tuberculosis == No AIDS = No }
{Sweating at Night == High, Temperature == Normal AIDS = No }
{Tuberculosis == No,Temperature == Normal AIDS = No }
{CD4 Cell Count > 500, Tuberculosis == No AIDS = No }
{CD4 Cell Count >500, Temperature == Normal AIDS = No }
{CD4 Cell Count > 500 , Sweating at Night == High, Temperature == Normal AIDS = No }
{CD4 Cell Count >500, Tuberculosis == No, Temperature == Normal}
{Sweating at Night == High, Tuberculosis == No, Temperature == Normal}
This clearly shows this algorithm generates minimal number of rules.
3.4 Weighted class association rule generation
Weighted Associative classifier construction is of two steps. In the first step, all the weighted
class association rules are generated based on weighted association rule mining technique. To
find the weighted association rule, weight is the important factor. HITS model is applied to the
dataset as in [17]. From this we can derive the weighted for each item.
Here weighted support and weighted confidence measures are used to evaluate the rule.
CD4 Cell
Count
Sweating at
Night
Tuberculosis (TB) Temperature AIDS
>500 High no Normal No
>500 High no High No
<200 High no Normal Yes
200 .. 500 Medium no Normal Yes
200 .. 500 Nil yes Normal Yes
200 .. 500 Nil yes High No
<200 Nil yes High Yes
200 .. 500 Medium no High No
>500 Medium no Normal No
>500 Nil yes Normal Yes
200 .. 500 Medium yes Normal Yes
>500 Medium yes High Yes
<200 Medium no High Yes
<200 High yes Normal Yes

7
Definition 1: The w-support of a rule X C is defined as
∑
→
=→
WeightHubTotal
CXWeightHub
CXWSup
)(
)(
(4)
Definition 2: The w-confidence of a rule X C is defined as
)(
)(
)(
XWConf
CXWConf
CXWConf
→
=→
(5)
Class association rule is said to significant if its weighted support is larger than the minimum
weighted support. Then the rules are ranked and the rules that satisfy certain threshold conditions
are used to construct the classifier. After rule ranking, only the high-ranking rules are chosen to
build a classifier and the rest are pruned.
3.5 Rule ranking and rule pruning
The minimum weighted support and minimum weighted confidence are user defined threshold
values. The itemset that has weighted support and weighted confidence above the threshold value
are called as frequent itemset and others are called as infrequent itemset which are pruned during
rule generation process. If all the frequent rules are used in the classifier then the accuracy of the
classifier would be high but the process of classification will be slow and time-consuming. So
rule ranking and rule pruning techniques are proposed to choose an optimal rule set. To apply rule
pruning, the generated rules are ranked based on several parameters and interestingness measures
such as weighted confidence, weighted support, lexicographical order of items etc. Initially the
rules are arranged based on their weighted confidence value. If two rules have the same value for
the weighted confidence measure then the rules are sorted based on their weighted support. If
both weighted confident and weighted support values are same for two rules then the sorting is
done based on their rule length. Even after considering weighted confidence, weighted support,
and cardinality measures, if there exists some rules with the same values for all three measures
then the rules are sorted based on its lexicographic order as in Lazy pruning [4] method.
3.6 Conflict Rule Reduction
Even after applying rule pruning strategies their may exist conflicting rules and redundant rules.
If there exists two rules r1 and r2 where
r1:X→Ci and r2:X→Cj
then r1 and r2 are said to be conflicting rules.
Let us consider a ruleset where, there exist two rules like
r1:X→Ci and r2:X ^ Y →Ci
then r1 and r2 are said to be redundant rule as they are semantically meaningless from a
classification viewpoint. In the gain based approach [7][25] confidence threshold was set greater
than 50%. This further enhances the ruleset as conflicting rules and redundant rules are avoided.
The remaining rules are used in the classifier.
3.7 Compact Weighted Associative Classification Algorithm
This section explains Compact weighted class association rule generation algorithm.

8
INPUT: DATA SET
OUTPUT: CLASS
I. Choose a non class informative attributet.
II. Generate one item- class association rules based on the selected attribute.
III. Calculate weighted support.
IV. If weighted support of item is greater than minimum weighted support then generate two
itemset and so forth
V. After generating all the itemset calculate weighted confidence for all itemset
Classifier Algorithm
I. Rank the rules based on Weighted Confidence, Weighted Support and Size of the rule
antecedent.
II. Classify the test dataset using these ruleset and obtain the classifier accuracy.
4. Experimental results
The proposed system was tested using benchmark datasets from the University of California at
Irvine Repository (UCI Repository) [6]. The datasets were preprocessed to convert to a general
format. A brief description about the datasets is presented in Table 3.The experiments were
carried out on a PC with Intel Core 2 Duo CPU with a clock rate of 1.60 Ghz and 2 GB of main
memory. Holdout approach [9] was used to randomly choose the training and testing dataset from
the dataset. The training dataset is used to construct a model for classification. After constructing
the classifier, the test dataset is used to estimate the classifier performance.
4.1 Accuracy Computation
Accuracy measures the ability of the classifier to correctly classify unlabeled data. It is the ratio
of the number of correctly classified data over the total number of given transactions in the test
dataset.
testsettheinobjectsofnumberTotal
classifiedcorrectlyobjectsofNumber
Accuracy =
(6)
The performance of the proposed compact weighted class association rule mining method was
evaluated by comparing it with the traditional associative classification algorithm (CBA) [13] and
the existing compact class association rule mining (GARC) [7]. Table 3 gives the dataset
description for various datasets.
TABLE 3 Dataset Description
Dataset Transactions Classes Items
Breast 699 2 18
Car 1728 4 25

9
Ecoli 336 8 34
Glass 214 7 20
Iris 150 3 12
Nursery 12960 5 27
Pima 768 2 36
Zoo 101 7 34
TABLE 4 Accuracy Comparison
DATASET IGmax Attribute CBA (conf-50 sup-1)
GARC
(Sup >1%,
Conf > 50)
CWAC
Breast 2 92.84 98.54 92.67
Car 6 90.51 88.50 78.33
Ecoli 6 80.36 69.27 86.14
Glass 8 61.68 61.90 76.19
Iris 4 93.33 94.60 94.60
Nursery 8 80.10 83.76 91.44
Pima 2 73.18 76.63 82.50
Zoo 3 83.18 85.14 88.00
Average 81.90 82.29 86.23
TABLE 5 Number of rules generated
DATASET
CBA (Sup>1%, conf >50%) GARC (Sup >1%
Conf>50%)
CWAC

10
Breast 1912 447 312
Car 993 313 298
Ecoli 1116 375 311
Glass 4859 906 850
Iris 194 109 74
Nursery 3897 3788 1095
Pima 1520 356 251
Zoo 75611 4367 2695
Average 11262.75 1332.63 735.75
TABLE 6 Accuracy Comparison
DATASET GARC (Sup > 1%, Conf >70)
CWAC
Breast 88.82 98.54
Car 81.57 79.02
Ecoli 70.48 87.95
Glass 77.14 76.19
Iris 94.60 96.00
Nursery 88.90 91.44
Pima 80.15 81.20
Zoo 86.00 86.00
Average 83.46 87.04
TABLE 7 Number of rules generated
DATASET GARC (Sup >1%, Conf > 70) CWAC

11
Breast 424 261
Car 365 184
Ecoli 250 224
Glass 643 593
Iris 110 73
Nursery 1197 762
Pima 285 138
Zoo 3897 678
Average 896.38 364.13
Table 4 shows the accuracy comparison for various dataset with minimum support of 1%,
minimum confidence is greater than 50%. The proposed algorithm has about +4.33 percent
improvements against the traditional associative classification and about +3.94 percent
improvements against the GARC respectively.
Table 5 shows number of rules generated. The traditional system has generated 15.31 times more
rules than the proposed system and the GARC algorithm generated about 1.81 times more rules
than the proposed system. If we consider the zoo dataset, it consists of only 101 transactions but it
has 18 attributes and 7 class attribute, this leads to generation of more number of rules. On the
other hand, the proposed system eliminates the un-necessary rules in the generation phase using
gain attribute as well as link based weight, so it generates only minimal number of rules.
For each dataset, the parameter of the algorithm such as minimum support and minimum
confidence varies to yield the best classifier. Obviously, the best setting for one dataset is
different from the other. A recent work [7, 25] suggests the best minimum support and minimum
confidence values as 1% and 70% respectively. Table 6 shows the accuracy comparison with
minimum support of 1%, minimum confidence is equal to 70%. The proposed compact weighted
associative classification method has about +3.58 percent improvements against the information
gain based class association rule mining. Table 7 shows number of rules generated for minimum
support 1%, minimum confidence of 70% and minimum chi square greater than of 3.84. The
traditional system has generated 2.46 times more rules than the proposed system.
The proposed system generates lesser number of rules, on the same time it improves the accuracy
of the system. So it is very easy to construct the classifier and to predict the new labels.
Conclusion
This paper proposes compact weighted class association rule mining method. The CWAC
algorithm aims to extract the weighted class association rules from the dataset. Weight is
computed using HITS algorithm, which does not require any preassigned weights. The proposed
CWAC algorithm chooses information gain attribute and generates all the rules based on that
attribute. So it generates compact ruleset. Experiment results shows that the proposed system not
only generated lesser number of rules but also increases the classification accuracy. The process

12
of rule generation and rule evaluation can be further enhanced by implementing other attribute
selection measures and rule evaluation strategies.
REFERENCES
1. Agarwal.R and Srikant.R (1994),” Fast algorithm for mining association rules in large data bases”,
Proceedings of the 20th
international conference on very Large Data Base(VLDB’94),
Santiago,chile, pp 487-499.
2. Agarwal.R, Imielinski.T and Swami.A (1993),” Mining Association rules between sets of items in
large databases”, proceedings of the 1993 ACM SIGMOD International conference on Management
of Data, Washington,DC,pp.207- 216 .
3. Baralis, E. and Torino, P (2004). “A lazy approach to pruning classification rules”. Proceedings of
the 2002 IEEE International Conference on Data Mining (ICDM’02), Maebashi City, Japan, pp.
35 – 42.
4. Baralis, E., Chiusano, S. and Graza, P (2008). “A lazy approach to associative classification”. IEEE
Transactions on Knowledge and Data Engineering, VOL. 20, NO. 2, February, pp 156 – 171.
5. Baralis, E., Chiusano, S. and Graza, P (2004). “On support thresholds in associative classification”.
In proceedings of the 2004 ACM Symposium on Applied Computing. Nicosia, Cyprus: ACM Press.
Pp 553–558.
6. Blake.C.L and Merz.C.J (1998), UCI Repository of machine learning databases.
7. Chen. G, Liu. H, Yu. L, Wei.Q and Zhang.X (2006), “A new approach to classification based on
association rule mining”, Decision Support Systems. pp- 674– 689.
8. Han, J. Pei, and Y. Yin (2000), “Mining Frequent Patterns without Candidate Generation,”
Proceedings of ACM SIGMOD, pp 1-12.
9. Han. J and Kamber.M (2001), “Data Mining: Concepts and Techniques”. New York: Morgan
Kaufmann Publishers.
10. Hong Yao, H.J Hamilton (2006),” Mining item set utilities from transaction data bases”,
IEEE Transactions on Data and Knowledge Engineering, volume 59, issue 3, pp.603-626.
11. Kleinberg.J.M (1999), “Authoritative Sources in a Hyperlinked Environment,” J. ACM, vol. 46, no.
5, pp.604-632.
12. Li, W., Han, J. & Pei, J. (2001) “CMAR: Accurate and efficient classification based on multiple-
class association rule”. In Proceedings of the International Conference on Data Mining (ICDM’01),
San Jose,CA, pp. 369–376.
13. Liu, B., Hsu, W. & Ma, Y (1998). “Integrating classification and association rule mining”. In
Proceedings of the International Conference on Knowledge Discovery and Data Mining. New York,
NY: AAAI Press,pp. 80–86.
14. Quinlan. J (1986). “Induction of decision trees”. Machine Learning, pp, 81–106.
15. Ramkumar.G.D, Ranka.S, and Tsur.S. (1998), “Weighted Association Rules: Model and
Algorithm,” Proc. ACM SIGKDD.

13
16. Sulaiman Khan.M, Muyeba,Frans Coenen. (2008), “A Weighted utility Framework for Mining
Association Rules”. In the Proceedings of second UKSIM European Symposium on Computer
Modeling and Simulation .
17. Sun Ke, Fengshan Bai (2008), ”Mining Weighted Association Rules without Preassigned Weights”
Proceedings of IEEE Transactions on Knowledge and Data Engineering.” VOL.20, No. 4, pp.489 –
495.
18. Syed Ibrahim S. P, Chandran K.R, Abinaya M.S (2011), “Integration of weighted association rule
mining and Classification” in the proceedings of National Conference on Recent Advances in
Computer Vision & Information Technology, March 7, 2011, pp 231-235.
19. Syed Ibrahim S. P, K. R. Chandran (2011), “Efficient Associative Classification Using Genetic
Network Programming” to be appeared in the International Journal of Computer Applications “,
September.
20. Syed Ibrahim. S.P, Chandran K.R, Abinaya.M.S (2011), “Compact Weighted Associative
Classification” in the IEEE International Conference on Recent Trends in Information Technology
(ICRTIT 2011), MIT, Anna University.Chennai, June 3-5, pp. 1099 – 1104.
21. Syed Ibrahim. S.P., K.R.Chandran, R.V.Nataraj (2011), “LLAC: Lazy Learning in Associative
Classification” in the Springer Lecture Series in Communications in Computer and Information
Science (CCIS), Advances in Communication and Computers, 190, Part I, PP. 631 – 638.
22. Syed Ibrahim.S.P, Chandran.K.R, Jabez Christopher.J (2011), “An Evolutionary Approach for
Ruleset Selection in a Class Based Associative Classifier” in the European Journal of Scientific
Research, ISSN 1450-216X, Vol.50 No.3, pp.422-429.
23. Tao. F, Murtagh, F., Farid, M.(2003), “Weighted Association Rule Mining Using Weighted Support
and Significance Framework.” In the proceedings of 9th ACM SIGKDD Conference on Knowledge
Discovery and Data Mining, Washington DC, pp. 661–666,
24. Yin.X and Han.J (2003), “CPAR: Classification Based on Predictive Association Rules,”
Proceedings of third SIAM Int’l Conf. Data Mining (SDM’03), May 2003.
25. Zhang. X, Chen.G, Wei. Q (2011), “Building a highly-compact and accurate associative classifier,
Applied Intelligence, Volume 34, Number 1, PP.74-86.
S. P. Syed Ibrahim received his BE in 2004 from Bharathidasan university and ME in 2006 from
Anna University, Chennai. He has been research scholar of Anna Univerisy – Coimbatore. He is
currently as Assistant professor in the department of Computer Science and Engineering at PSG
College of Technology, Coimbatore. He has published more than 15 research papers in various
journals and conferences. His research interest includes data mining, storage management.
K.R. Chandran received his BE in 1982 and ME in 1986 from PSG College of Technology and
PhD in 2006 from Bharathiar university, Coimbatore. He is currently professor in the department
of Information Technology at PSG College of Technology, Coimbatore. He is a Fellow member
in Institution of Engineers (India) and life member of ISTE, ACS, and The Textile Association
(INDIA). He has published more than 55research papers in various journals and conferences.
His research interest includes software engineering, computer networks, data mining, system
modeling and simulation.

COMPACT WEIGHTED CLASS ASSOCIATION RULE MINING USING INFORMATION GAIN

More Related Content

What's hot (18)

Similar to COMPACT WEIGHTED CLASS ASSOCIATION RULE MINING USING INFORMATION GAIN (20)

Recently uploaded (20)

COMPACT WEIGHTED CLASS ASSOCIATION RULE MINING USING INFORMATION GAIN