3.[18 22]hybrid association rule mining using ac tree

Journal of Information Engineering and Applications www.iiste.org
ISSN 2224-5758 (print) ISSN 2224-896X (online)
Vol 1, No.2, 2011

Hybrid Association Rule Mining using AC Tree
Preethi Kolluru Ramanaiah1*
1. Business Technology Analyst, Deloitte, Bangalore, India
* E-mail: preethiram4@gmail.com

Abstract
In recent years, discovery of association rules among item sets in large database became popular. It
gains its attention on research areas. Several association rule mining algorithms were developed for
mining frequent item set. In this papers, a new hybrid algorithm for mining multilevel association rules
called AC Tree i.e., AprioriCOFI tree was developed. This algorithm helps in mining association rules
at multiple concept levels. The proposed algorithm works faster compared to traditional association
rule mining algorithm and it is efficient in mining rules from large text documents.
Keywords: Association rules, Apriori, FP tree, COFI tree, Concept hierarchy.

1. Introduction
At present, association analysis [8] was widely used in many industries. This has significant usage in
market-basket analysis. The main aim of association analysis is the discovery of association rules of the
form attribute-value. The 2 major sub problems in mining association rules are:
• Finding all frequent item sets that appear more than minimum support a
• Generating association rules using these frequent item sets.
A number of algorithms for mining frequent item sets have been proposed after Agarwals [7] first work
of deriving categorical association rules from transactional databases. And then, many new algorithms
have been developed for efficient mining. Now, these existing algorithms can be categorized in two
classes called candidate-generate approach and pattern growth approach. The best suited example for
candidate generate approach is Apriori[6] and pattern growth approach is Fp Tree[7].

1.1 Candidate-generate Approach
The candidate-generate approach is most commonly used and most widely popular approach for
mining data from transactional databases. The two main steps in this method are join and prune[8,7].
Here k-frequent itemsets are joined to form (k+1) th candidate itemsets. In each and every iterations, the
database is scanned o discard itemsets that do not meet minimum support. This algorithm is called as
Apriori algorithm. It achieves good reduction on the size of candidate sets. However, it takes many
scans of database to check the candidates support and this suffer from larger overhead of I/O scans.
This is consider to be the main disadvantage of Apriori algorithm on mining large text documents.

1.2 Pattern growth Approach
The next set of algorithms for mining association rules are pattern growth approach [6]. Over the past
few years, many pattern-growth approaches have been developed. Some of the most popular algorithms
are Fp Growth, Tree-projection, H-mine, COFI, BOMO, etc. The main aim of pattern-growth
approach is the use of tree to store database instead of generating candidates. Thus this method proves
to be much faster compared to candidate-generate approach. However, it suffers from the problem of
memory overhead for mining data from large database. But for smaller data set, this approach is
consider to be most suited approach.

2. Multilevel Association Rule Mining
Important development in pattern-growth approach is multilevel association rule mining which
involves items at different levels of abstraction. From [7], this method involves 2 main stage

18 | P a g e
www.iiste.org

Vol 1, No.2, 2011

• construction of a modified frequent pattern tree
• Repetitive building of small data structures.
Here, association rules are generated at multiple level using the frequent patterns at related concept
level. The best suited example for multilevel association rule mining algorithm is COFI tree called as
Co-Occurrence Frequent Item tree.

2.1 COFI tree
COFI tree[3, 4] is mainly used for finding more interesting patterns hidden at multiple concept level. It
deals with the construction of relatively small trees for each frequent item in the header table of Fp tree.
Thus the main important step in COFI tree first involves construction of Fp header table. Then pruning
is done to remove all non-frequent itemsets with respect to minimum support threshold. Then at each
level, it eliminates all non-frequent itemsets that will not participate in subsequent levels. This is
similar to Apriori property [5]. The elimination at each of the levels is to find all frequent patterns with
respect to one frequent item. The process is carried out for each and every element in the header table
of Fp tree. Subsequently, the join process is carried out at every level to generate candidate like itemset
for performing next iteration. Since, it requires the construction of Fp tree for each and every frequent
item in the header table, the memory requirement is more. Hence it is not scalable for larger dataset.

3. Hybrid AC Tree Algorithm
In order to efficiently mine frequent patterns from larger data set, the new algorithm called AC tree is
developed. It is hybrid algorithm that combines Apriori and COFI tree and FP tree. The main aim of
the project is to overcome the disadvantage of those traditional algorithms and make it suitable for
mining larger dataset efficiently. Figure 3.1 shows the complete architecture.
3.1 AC tree
The AC tree mainly involves 2 steps. In the first step, the dataset is scanned one time to find out
frequent 1-itemsets using Apriori algorithm. The main aim of this step is to increase the efficiency in
doing text mining. The text document will have many words which are irrelevant for analysis. Thus as
a initial step, data preprocessing is carried out to remove all stop words. Then, using Apriori algorithm,
1- frequent itemset is discovered. The idea behind discovering 1 frequent itemset and not more is, this
is enough to give sufficient result which helps to carry out further steps. In the second step, Fp header
table is constructed using those 1- frequent itemset. For each and every item in the Fp header table, a
tree is constructed which contains all the transactions that particular element is presented. Only
considering those transactions where particular element , say x, is presented. Fp tree is constructed for
element x. Likewise, the process repeated for each and every element. Then frequent patterns are found
out at every level of iteration. Pruning is done based on its minimum support count. Thus frequent
pattern is mined at each and every level hence it is called as multilevel association rule mining
algorithm. Since it is multilevel rule mining algorithm, the number of iterations depends on the type of
dataset involved. The mining process is recursive until there exist 1 frequent itemset at ith level. This is
based on assumption that, if number of patterns in nth level is 1, then there exist no frequent itemset in
(n+1)th level. The algorithm is as follows.

Algorithm
Step 1: Scan the data set once In this process the given text dataset is scanned fully and the word
frequency is noted for each and every word
Step 2: generate all 1-frequent item sets that satisfy minimum support count using Apriori
Step 3: sort the items in candidate item set in decreasing order of frequency
step 4: now choose the lowest frequent item set 'x' and make note of all transactions that contains the
item x
step 5: construct a Fp tree for each and every item set x along with the transactions that contain item x
step 6: find the frequent patterns at each level of construction of Fp tree
step 7: repeat step 4 until sub conditional Fp tree is constructed for each and every element

19 | P a g e
www.iiste.org

Vol 1, No.2, 2011

step 8: now perform join operation of all conditional sub Fp trees
step 9: at each and every level of join operations, find the frequent item sets.
Thus, from the above algorithm, we mine frequent items at multiple levels.

Figure 3.1 Architecture of AC tree

4.Experimental Results
The proposed algorithm was implement on java. The dataset used for analysis is 20 Newsgroup dataset.
In order to verify its performance, the AC tree algorithm was compared with Apriori , Fp tree and
APFT[8] algorithm. Experiments shows that AC tree outperforms the above traditional algorithms. It
works faster and helps to discover more interesting patterns at multiple level. This algorithm is
efficiently used for mining a large corpus text documents.

20 | P a g e
www.iiste.org

Vol 1, No.2, 2011

Figure 4.1 Comparisons between AprioriGrowth and AprioriCOFI(AC Tree)

The above chart i.e. Figure 4.1 shows the comparison between AprioriGrowth, AprioriCOFI (AC) tree
and Fp tree. In order to find all nth level item sets, AC tree yields a better result. And also time taken
for mining frequent item sets are comparatively minimum in AC tree than Apriori growth algorithm.
The number of patterns mined is more interesting because it helps to identify the hidden interesting
patterns at multiple levels. Thus AC tree works better than Apriori growth and Fp tree.

References
[1] Qihua Lan, Defu Zhang, (2009) “A new algorithm for frequent itemsets mining based on Apriori
and FP-tree”, Ref no. 978-0-7695-3571-5/09 2009 IEEE, global congress on Intelligent systems, pp.
1099-1102.
[2] Mohammad el-hajj, Omar R.Zaiane, (2005) “COFI Tree mining: a new approach to pattern growth
with reduced candidacy generation” International Journal of Business Intelligence and Data Mining
Volume 1 Issue 1, July 2005, pp. 88-106.
[3] Sze-chung Ngan, Tsang Lam, Raymond chi-wing wong and Ada wait-chee fu (2005) “ mining N
most interesting itemsets without support threshold by COFI -tree” , Int. J.Business Intelligence and
Data mining, vol. 1, No. 1, 2005, pp. 348-354.
[4] Mohammad El-Haii and Osmar R, (2003) “COFI-tree mining: A new approach to pattern Growth
within the context of interactive mining” In Proc. 2003 Int'1 Conf. On Data mining and Knowledge
Discoverey (ACM SIGKDD), August 2003 pp. 109-118
[5] Yongwppk yoon and Gary geunbae lee, (2003) “practical application of associative classifier for
document classification”, BK21 program of Korea Ministry of education and grant number R01-2003-
000-10181-0, pp. 331-339.

21 | P a g e
www.iiste.org

Vol 1, No.2, 2011

[6] J.Han, J.Pei and Y.Yin.(2000), “Mining frequent patterns without candidate Generation”,
Proceeding of ACM SIGMOD International Conference Management of Data, 2000, pp. 1-12.
[7] Agrawal R and R.Srikant, (1994) “fast algorithms for mining association rules”, In VLDBY94,
ISBN:1-55860-523-1, pp.487-499.
[8] Agrawal R, T.Imielinski and A.Swami (1993). “Mining association rules between set of items in
large databases” In Proc.1993 ACM-SIGMOD Int conf. Management of Data, Washington.D.C, may
1993, pp. 207-216.

22 | P a g e
www.iiste.org

3.[18 22]hybrid association rule mining using ac tree

More Related Content

What's hot (17)

Similar to 3.[18 22]hybrid association rule mining using ac tree (20)

More from Alexander Decker (20)

Recently uploaded (20)

3.[18 22]hybrid association rule mining using ac tree