Pratima Gautam et. al.

/ (IJCSE) International Journal on Computer Science and Engineering

Vol. 02, No. 03, 2010, 746-752

A Fast Algorithm for Mining Multilevel
Association Rule Based on Boolean Matrix
1
Pratima Gautam
2
K. R. Pardasani
Department of computer Applications Department of Mathematics
MANIT MANIT
Bhopal (M.P.) Bhopal (M.P.)
Abstract - In this paper an algorithm is proposed for
mining multilevel association rules. A Boolean
Matrix based approach has been employed to
discover frequent itemsets, the item forming a rule
come from different levels. It adopts Boolean
relational calculus to discover maximum frequent
itemsets at lower level. When using this algorithm
first time, it scans the database once and will
generate the association rules. Apriori property is
used in prune the item sets. It is not necessary to
scan the database again; it uses Boolean logical
operation to generate the multilevel association rules
and also use top-down progressive deepening
method.

Keywords - association rules, data mining, fuzziness,
multilevel rules.
I. INTRODUCTION
Data mining, or the efficient discovery of
interesting patterns from large collections of data,
has been recognized as an important area of
database research. The most commonly sought
patterns are association rules as introduced in [4].
Association rule mining is an important data
mining technique to generate correlation and
association rule. The problem of mining
association rules could be decomposed into two
sub problems, the mining of large itemsets (i.e.
frequent itemsets) and the generation of
association rules[ 1] [3]. An association rule is an
implication of the form A B, where A_I, B_I,
and AB= | . The rule A B holds in the
transaction set D, with support s, where s is the
percentage of transactions in D that contain A B
(i.e., the union of sets A and B, or say, both A and
B). This is taken to be the probability,
P(AB).The rule A B has confidence c in the
transaction set D, where c is the percentage of
transactions in D containing A that also contain B.
This is taken to be the conditional probability,
P(B|A). That is,
) ( ) ( B A P B A Support =
) | ( ) ( A B P B A Confidence =


Rules that satisfy both a minimum support
threshold (min sup) and a minimum confidence
threshold (min conf) are called strong [14]. A set
of items is referred to as an itemset. An itemset
that contains k items is a k-itemset. The set of
{computer, laser printer} is a 2-itemset. The
occurrence frequency of an itemset is the number
of transactions that contain the itemset. This is
also known, simply, as the frequency, support
count, or count of the itemset [13]. Note that the
itemset support defined in the equation given
below is sometimes referred to as relative support,
whereas the occurrence frequency is called the
absolute support. If the relative support of an
itemset I satisfies a prespecified minimum support
threshold (i.e., the absolute support of I satisfies
the corresponding minimum support count
threshold), then I is a frequent itemset [2]. The set
of frequent k-itemsets is commonly denoted by L
k
.
) (
) (
) | ( ) (
A Support
B A Support
A B P B A Confidence

= =
The problem of mining association rules can be
reduced to that of mining frequent itemsets [10].
Association rule mining can be viewed as a two-
step process:
1. Find all frequent itemsets: By definition, each
of these itemsets will occur at least as frequently

as a predetermined minimum support count, min
sup.
2. Generate strong association rules from the
frequent itemsets: By definition, these rules must
satisfy minimum support and minimum
confidence.
It is difficult to find strong and interesting
associations among data items at the primitive
levels of abstraction due to the paucity of data.
ISSN : 0975-3397 746
Pratima Gautam et. al. / (IJCSE) International Journal on Computer Science and Engineering
Vol. 02, No. 03, 2010, 746-752

However, many strong associations discovered at
rather high concept levels are common sense
knowledge. Therefore, a mining system with the
capabilities to mine association rules at multiple
levels of abstraction and traverse easily among
different abstraction spaces is more desirable like
Han et al. in [15] and Rajkumar et al. in [8]
indicate.
We are using multilevel association rule and
Boolean association rule in our algorithm called
MLBM. Boolean association rule mining is used
more widely than other kinds of association rule
mining [9]. This algorithm transforms a
transaction database into a Boolean matrix stored
in bits. Meanwhile it uses the Boolean vector
relational calculus method to discover frequent
itemsets. We use the fast and simple and
calculus in the Boolean matrix to replace the
calculations and complicated transactions that
deal with large numbers of itemsets [5]. This
algorithm is more effective than the Apriori-like
algorithms. This algorithm is also used
progressive deepening method. The method first
finds frequent data items at the top most level and
then progressively deepens the mining process
into their frequent descendants at lower concept
levels [7]. This method is using concept of
reduced support and refine the transaction able at
each level.

II. APORIORI ALGORITHM
The key of mining association rules is to set an
appropriate support and confidence values to find
frequent itemset. The well-known algorithm,
Apriori, exploits the following property: If an
itemset is frequent, so are all its subsets [1].
Apriori employs an iterative approach known as
level wise search, where k -itemsets are used to
explore k 1-itemsets. First, the set of frequent 1-
itemsets is found. This is denoted as L
1
. L
1
is used
to find L
2
, the frequent 2-itemsets, which is used
to find L
3
, and so on, until no more frequent k -
itemsets can be found. The finding of each L
k

requires one full scan of the database. Throughout

the level-wise generation of frequent itemsets, an
important anti-monotone heuristic is being used to
reduce the search space [16].
III. MULTI LEVEL ASSOCIATION
RULE MINING
We can mine multilevel association rules
efficiently using concept hierarchies, which
defines a sequence of mappings from a set of low-
level concepts to higher-level, more general
concepts [6] [17]. Data can be generalized by
replacing low-level concepts within the data by
their higher-level concepts or ancestors from a
concept hierarchy. In a concept hierarchy, which
is represented as a tree with the root as D i.e.,
Task-relevant data. The popular area of
application for multi level association is market
basket analysis [8] [11], which studies the buying
habits of customers by searching for sets of items
that are frequently, purchased together which was
presented in terms of concept hierarchy shown
below. Each node indicates an item or item set
that has been examined. There are various
approaches for finding frequent item sets at any
level of abstraction. Some of the methods which
are in use are using uniform minimum support
for all levels, using reduced minimum support at
low levels, level-by-level independent.
Multi-level databases use hierarchy-information
encoded transaction table instead of the original
transaction table [7]. This is useful when we are
interested in only a portion of the transaction
database such as food, instead of all the items.
This way we can first collect the relevant set of
data and then work repeatedly on the task-
relevant set. Thus in the transaction table each
item is encoded as a sequence of digits.
Example: encoded as a sequence of digits in the
transaction table T [1]. For example, the item `2
percent foremost milk' is encoded as `112' in
which the first digit, `1', represents `milk' at level-
1, the second, `1', for `2 percent (milk)' at level-2,
and the third, `2', for the brand `Foremost' at
level-3. Similar to [15], repeated items (i.e., items
with the same encoding) at any level will be
treated as one item in one transaction.



Fig1. The taxonomy for the relevant data items

IV. A MULTILEVEL ALGORITHM BASED
ON BOOLEAN MATRIX (MLBM)
We propose a new multilevel association
algorithm. The section is organized as follows: the
correlative definition and proposition, an
introduction to the MLBM algorithm details, and
a description of a sample execution of the MLBM
algorithm.
ISSN : 0975-3397 747
Pratima Gautam et. al. / (IJCSE) International Journal on Computer Science and Engineering
Vol. 02, No. 03, 2010, 746-752

A. Definition and proposition
Association Rules
Definition 1: Let I = {i
1
, i
2
, i
3
,. i
n
} be a set of
items. D is a database of transactions. Each
transaction T is a set of items and has an identifier
called TID. Each T_I. [9]
Definition 2: Association rule is the implication
of the form A B, where A and B are itemsets
which satisfies A_I, B_I and A

B =| .
Definition 3: The strength of an association rule
can be measured in terms of its Support and
Confidence. Rule A B is true in D with a
support (denoted by sup) and a confidence
(denoted by conf), where A and B are set of
items. Support sup is a percentage of
transactions including both A and B (A B) in
transaction sets D. Confidence conf is a
percentage of transactions including both A and B
(AB) in transactions that contain A.[12]
Sup=P (AB)/|D|, conf =P (B|A) P (AB)/P
(A)
Definition 4: Boolean Matrix: is a matrix with
element 0 or 1.
Definition 5: The Boolean AND operation is
defined as follows:
0.0=0 0.1=0 1.0=0 1.1=1
Where logical implication is denoted by . or
AND. If we write C=A.B, then C can be
determined by listing all possible combinations of
A and B. Truth table for logical AND will be:
TABLE I . AND OPERATOR
A B C=A.B
0 0 0
0 1 0
1 0 0
1 1 1
Definition 6: The Boolean AND calculus is
carried out to an arbitrary k columns vector of the
Boolean matrix; the sum of 1 of the operation
result is called k- support of the k columns vector.
Proposition 1: If the sum of 1 in a column
vector A
i
is less than min_sup_num, it is not
necessary that A
i
will attend calculus of the next
level supports.
Rationale: According to the principle of the
Boolean AND calculus, the result is 1 when the
value of all vector elements (in a record) is 1 [5].
Proposition 2: Itemset A is a k-itemsets (each
item belongs to different level); |LK-1(j)| presents
the number of values in a level j in all frequent
(k-1)-itemsets of the frequent set LK-1. There is
an item j in X. If |LK-1(j)| is smaller than k-1,
itemset X is not a frequent itemset[5] .

b. Algorithm Details (MLBM)
The algorithm consists of following steps:
Step-1:
Encode taxonomy using a sequence of numbers
and the symbol *, with the lth number
representing the branch number of a certain item
at levels.
Step-2:
Set H = 1, where H is used to store the level
number being processed whereas H e{1, 2, 3}
(as we consider up to 3-levels of hierarchies).
Step-3:
Transforming the transaction database into the
Boolean matrix.
Step-4:
Set user defines minimum support on current
level.
Step-5
Generating the set of frequent 1-itemset L
1
at level
1.
Step-5:
Pruning the Boolean matrix
Step-6:
Perform AND operations to generate 2-itemsets
and 3- itemset at level 1.
Step-7:

Generate H +1; (Increment H value by 1; i.e., H =
2) itemset from L
k
and go to step-4 (for repeating
the whole processing for next level).

c. Transforming the transaction database
into the Boolean matrix
The mined transaction database is D, with D
having m transactions and n items. Let T= {T1,
T2,, Tm} be the set of transactions and I= {I
1
,
I
2
,,I
n
) be the set of items. We set up a Boolean
matrix Am*n, which has m rows and n columns.
Scanning the transaction database D, if item Ij is
in transaction Ti , where 1jn the element value
of A
i
is 1, otherwise the value of Ij is 0.

d. Generating the set of frequent 1-itemset
L
1
The Boolean matrix Am*n is scanned and support
numbers of all items are computed. The support
number Ij.supth of item Ij is the number of 1s in
the jth column of the Boolean matrix Am*n. If
Ij.supth is smaller than the user define minimum
support number minsupth, itemset {Ij} is not a
frequent 1-itemset and the jth column of the
Boolean matrix Am*n will be deleted from Am*n.

e. Pruning the Boolean matrix
Pruning the Boolean matrix means deleting some
columns from it. This is described in detail as: Let
ISSN : 0975-3397 748
Pratima Gautam et. al. / (IJCSE) International Journal on Computer Science and Engineering
Vol. 02, No. 03, 2010, 746-752

I be the set of all items in the frequent set LK-1,
where k>2. Compute all |LK-1(j)| where jeI, and
delete the column of correspondence item j if |LK-
1(j)| is smaller than min_sup_num.

f. Generating the set of frequent k-itemsets
L
k

Frequent k-itemsets are discovered by AND
relational calculus, which is carried out for the k-
vectors combination. If the Boolean matrix Ap*q
has q columns where 2<qn and min_sup_num is
hpm, (C
q
)
k
, combinations of k-vectors will be
produced. The AND relational calculus is for each
combination of k-vectors. If the sum of elements
values in the AND calculation result is not
smaller than the minimum support number
min_sup_num, the k-itemsets corresponding to
this combination of k-vectors are the frequent k-
itemsets and are added to the set of frequent k-
itemsets L
k
.



5. An Illustrative Example:
An illustrative example is given to understand
well the concept of the proposed algorithm and
how the process of the generating multilevel

association rule mining is performed step by step.
The process is started from a given transactional
database as shown in Table 1[a].
Table 1[a]
Trans_ID List of items
T1 111, 212, 311
T2 111, 222, 311, 411, 511
T3 111, 222, 411
T4 121, 321, 422,521
T5 111, 222, 311, 411
T6 222, 311, 422

Table.1 [a] Transaction data of the transaction
database D.

Table1 [b]
Codes of item name

Code Description
1** Milk
2** Bread
3** Cookies
4** Fruit
5** Drink
11* 2%
12* Skimmed
21* White
22* Wheat
31* Black Tea
32* Green Tea
41* Apple
42* Orange
51* Cola
52 * Drink Prigat
111 Milk 2% Amul
121 Milk Skimmed Anik
211 Bread White Wonder
222 Bread wheat Foremost
311 Cookies black Tea Nestle
321 Cookies Green Tea Linton
411 Fruit Apple red Delicious
422 Fruit Orange Valencla
511 Drink cola Coca
522 Drink Prigat pepsi

The transaction database D is transformed into the
Boolean matrix A6*5:






Fig. 2 The Boolean matrix A6*5
We execute the MLBM algorithm at level-1.
Therefore minimum support number = 3.0


We compute the sum of the element values of
each column in the Boolean matrix A6*5 and the
set of frequent 1-itemset is:
{{1**}, {2**}, {3**}, {4**}}
In pruning the Boolean matrix A6*5 .The fifth
column of the Boolean matrix A6*5 is deleted
because the support number of item 5** is smaller
than the minimum support number. Finally, the
Boolean matrix A6*4 is generated.

1* * 2** 3** 4**

(
(
(
(
(
(
(
(

1 1 1 0
1 1 1 1
1 1 0 1
1 0 1 1
1 1 1 1
0 1 1 1


T1

T2
T4
T3
T5
T6
ISSN : 0975-3397 749
Pratima Gautam et. al. / (IJCSE) International Journal on Computer Science and Engineering
Vol. 02, No. 03, 2010, 746-752

We perform the AND operation to generate 2-
itemset at level-1.And now matrix is A6*6.
The possible 2-itemsets are: (1**. 2**),
(1** . 3**), (1**. 4**), (2** . 3**), (2**. 4**),
(3** . 4**)



We compute the sum of the element values of
each column in the Boolean matrix A6*6 and all 2-
itemset considered for further process because
their support numbers are greater than the
minimum support number. Again we perform
ANDoperation to generate 3-itemset and finally
matrix is A6*4.
The possible 3-itemsets are: (1**. 2**. 3**)
(1**. 2**. 4**)(1**. 3**. 4**) (2** . 3**. 4**)
We compute the sum of the element values of
each column in the Boolean matrix A6*4 and all 3-
itemset considered for further process because
their support numbers are greater than the
minimum support number and we go to next level.
Level-2
Minimum_support = 2.0
1-itemset

* * 1

* * 2

* * 3

* * 4



We compute the sum of the element values of
each column in the Boolean matrix A6*8 and 12*,
21*, 32* column are deleted because their support
numbers are smaller than the minimum support
number. Again perform AND operation to
generate 2-itemset at level 2.Finally, the Boolean
matrix A6*9is generated.

The possible 2-itemsets are: (11*. 22*)
(11* . 31*) (11*. 41*) (11* . 42*) (22* . 31*)
(22*. 41*) (22* . 42*) (31* . 41*) (31* . 42*)



We compute the sum of the element values of
each column in the Boolean matrix A6*9.
(11*. 42*)(22*. 42*)(31*. 42*) column are
deleted because their support numbers are smaller
than the minimum support number. Again we
perform AND operation to generate 3-itemset
and finally matrix is A6*3 generated.

The possible 3-itemsets are: (11*. 22*. 31*)
(11*. 22*. 41*) (22*. 31*. 41*)

(11* . 22* . 31*) (11* . 22* . 41*) (22* . 31* . 41)




ISSN : 0975-3397 750
Pratima Gautam et. al. / (IJCSE) International Journal on Computer Science and Engineering
Vol. 02, No. 03, 2010, 746-752


We compute the sum of the element values of each
column in the Boolean matrix A6*3 and all 3itemset
considered for further process because their support
numbers are greater than the minimum support
number and we go to next level. Finally matrix is
A6*4.
Level-3
Minimum_support = 2.0
1-itemset


We compute the sum of the element values of


each column in the Boolean matrix A6*3 and all 3-
itemset considered for further process because
their support numbers are greater than the minimum
support number. Now we perform AND operation
on 1-itemset at level-3 and generated 2-itemset,
finally matrix is A6*6.

The possible 2-itemsets are: (111. 222) (111 . 311)
(111 . 411) (222. 311) (222. 411) (311. 411)




We compute the sum of the element values of
each column in the Boolean matrix A6*6 and all 2-
itemset considered for further process because
their support numbers are greater than the
minimum support number and Now we perform
AND operation on 2-itemset to generate 3-
itemset at level-3. Finally matrix is A6*3 generated.
(111 . 222 . 311) (111 . 222 . 411) (222 . 311 . 411)





According to step-3, the MLBM algorithm
is terminated because there are maximum
frequent itemset find at lower level(level3).
CONCLUSION
In this paper, a multilevel association rule mining
algorithm based on the Boolean matrix (MLBM)
is proposed. The main features of this algorithm
are that it only scans the transaction database
once, it does not produce itemsets, and it adopts
the Boolean vector relational calculus to
discover frequent itemset. In addition, it stores all
transaction data in bits, so it needs less memory
space and can be applied to mining large
transaction databases.

Reference:
[1] R. Agrawal, T. Imielinski, and A. Swami, Mining
association rules between sets of items in large databases,
Proceedings of the ACM SIGMOD Conference on
Management of data, pp. 207-216, 1993.
[2]R. Agrawal and R. Srikant, Fast algorithms for mining
association rules, In proceeding of the VLDB
Conference,1994.
[3] H. Mannila, H. Toivonen, and A, Verkamo. Efficient
algorithm for discovering association rules, AAA1 Workshop
on Knowledge Discovery in Databases.
[4]Jiawei Han, Micheline Kamber, Data Mining Concepts
and Techniques, Higher Education Press 2001.
[5] Hunbing Liu and Baishen wang, An association Rule
Mining Algorithm Based On a Boolean Matrix, Data
Science Journal, Vol-6, Supplement 9, S559-563, September
2007.
[6] R.S Thakur, R.C. Jain, K.R.Pardasani, "Fast Algorithm for
Mining Multilevel Association Rule Mining," Journal of
Computer Science, Vol-1, pp. 76-81, 2007 .
[7] Ha and Y. Fu, Mining Multiple-Level Association Rules
in Large Databases, IEEE TKDE. Vol-1, pp. 798-805, 1999 .


T6






T2
ISSN : 0975-3397 751
Pratima Gautam et. al. / (IJCSE) International Journal on Computer Science and Engineering
Vol. 02, No. 03, 2010, 746-752

[8] N.Rajkumar, M.R.Karthik, S.N.Sivana and
S.N.Sivanandamndam, "Fast Algorithm for Mining Multilevel
Association Rules, IEEE, Vol-2, pp. 688- 692, 2003.
[9] Maurice Houtsma and A , Swam I, Set-oriented mining
of association rules, Research Report RJ 9567, IBM
Almaden Research Center[C], San Jose, Califomia:[s.n.] 1993
[10] A.K.H. Tung, H. Lu, J. Han and L, Feng, Efficient
mining of intertransaction association rules, IEEE Trans.on
Knowledge and Data Engineering, vol-15, no. 1, pp.43-56,
Jan./Feb. 2003.
[11] S. Brin, R. Motwani, J. D. Ullman and S. Tsur, Dynamic
itemset counting and implication rules for market basket data,
ACM SIGMOD International Conference on Management of
Data, pp. 255264, May 1997.
[12] Anjna Pandey and K. R. Pardasani,Rough Set Model for
Discovering Hybrid Dimensional Association Rules,
International Journal of Computer Science and Network
Security, Vol -9, no.6, pp.159-164,2009.
[13] Ravindra Patel, D. K. Swami, K. R. Pardasani,
Lattice Based Algorithm for Incremental Mining of
Association Rules, International Journal of Theoretical and
Applied Computer Sciences, Vol- 1, pp. 119128, 2006.
[14] Jun Gao ,Realization of a New Association Rule
Mining Algorithm, IEEE DOI, 2007.
[15] Scott Fortin and Ling Liu,"An object-oriented approach to
multi-level association rule mining," Proceedings of the fifth
international conference on Information and knowledge
management, pp.65 72 1996.
[16] Neelu Khare , Neeru Adlakha and K. R. Pardasani
Karnaugh Map Model for Mining Association Rules in Large
Databases, (IJCNS) International Journal of Computer and
Network Security,Vol- 1, No. 1, October 2009
[17] Pratima Gautam Neelu Khare and K. R. Pardasani, A
model for mining multilevel fuzzy association rule in
database, Journal of computing, vol- 2, issue- 1, pp. 58-68
January 2010




ISSN : 0975-3397 752