International Journal on Recent and Innovation Trends in Computing and Communication

Volume: 4 Issue: 3

ISSN: 2321-8169
75 - 78

_______________________________________________________________________________________

Comparative Study of Different Improvements of Apriori Algorithm

Anneshya Ghosh

Ambar Dutta

Department of Computer Science

Birla Institute of Technology, Mesra, Kolkata Campus
Kolkata ,India
anneshya.ghosh@gmail.com

Department of Computer Science

Birla Institute of Technology, Mesra, Kolkata Campus
Kolkata ,India
adutta@bitmesra.ac.in

Abstract Data mining is a process of finding out the most frequent patterns from a large set of dataset. Association rule mining is an important
technique in data mining. Apriori algorithm is the most basic, popular and simplest algorithm for finding out this frequent patterns. Still being
one of the simplest algorithms for association rule mining, it has certain limitations. In the literature, researchers have proposed several
improvements of Apriori algorithms. This paper provides the comparative study of some of these improved version of Apriori algorithm with
respect to traditional Apriori algorithm.
Keywords- Data Mining, Apriori algorithm, Frequent Itemset, Support, Dataset.

__________________________________________________*****_________________________________________________
I. INTRODUCTION
Data Mining [1] is a process where information or
knowledge is being extracted or mined from a large set of
dataset. Here we mine the sequential patterns that are present in
the large databases. Sequential Pattern Mining was first
introduced by Agarwal and Srikant in the year 1995. Sequence
patterns are those set of data which occurs in a specific order
that is sequentially among all the data patterns in a given set.
And finding out of these patterns which occurs sequentially out
of all the other patterns is sequential pattern mining. An
example of sequential pattern is as follows; suppose a customer
buys a laptop then it is more likely that the customer will buy a
mouse then antivirus and a printer after it sequentially. Some
terms which are constantly being used here are item-set,
support and confidence. Let there be a set of items L = {l1, l2,
} and a sub set of these items is knows as item-set. For a
given database D, support of an item, let be X is defined as the
ratio of the number of sequences in the database which contain
the item X to the total number of sequences in the database.
And, for a given database D, confidence of an sequence that
contains X as well as Y is defined as the percentage of the
number of sequences that contains X as well as Y to the
number of sequences which contains X. Mining of sequential
patterns can be classified into three different categories, they
are as 1. Mining based on candidate generation (example,
Apriori algorithm), 2. Mining without the involvement of any
candidate generation (example, FP-Growth Tree algorithm) and
3. Mining item sets which have vertical format (example,
ECLAT algorithm).
Mining of sequential patterns can be classified into three
different categories, they are as 1. Mining based on candidate
generation (example, Apriori algorithm), 2. Mining without the
involvement of any candidate generation (example, FP-Growth
Tree algorithm) and 3. Mining item sets which have vertical
format (example, ECLAT algorithm).Apriori algorithm is the
algorithm which involves Candidate generation. According to
this algorithm, first the 1-itemsets are found then the database
is scanned to find the support count. The itemsets with support
count less than minimum support count are discarded. The
resultant itemsets are then used to find the frequent 2-itemsets
in the same process. Likewise we find all the (k+1)-itemsets
from the frequent k-itemsets, until no more frequent itemsets
can be found out. In FP-Growth tree algorithm [2], candidate
keys are not generated and database is scanned for two times

only. It uses a tree like structure to store the database and uses a
divide and conquer method. And in ECLAT algorithm [2],
depth first search method is used. In first scan of the database a
TID (Transcation_Id) list is given to each single item. k+1
Itemset are then generated from the k itemset using apriori
property and depth first search method. (k+1)-Itemset are then
generated by taking the intersection of the TID-set of frequent
k-Itemset. This process is continued, until no more candidates
Itemset can be found.
In this paper, Apriori algorithm is taken into consideration.
In section II, a detailed description of Apriori algorithm is
provided along with its limitations. In section III, some of the
existing improvements of Apriori algorithm are discussed with
examples. In section IV comparisons between the original and
the existing improved apriori algorithm is shown. Finally,
conclusion is derived in section V.
II.

APRIORI ALGORITHM

A. Description
The first and the most basic algorithm which was developed
to find out the sequential patterns from a database was the
Apriori algorithm [3]. This algorithm involves candidate
generation and was first proposed by R. Agarwal and R.
Srikant in the year 1994. In Apriori algorithm, we first scan the
original database and find out the support count of each of the
individual items. And discard those items whose support count
is less than the minimum support count. The resultant item set
is then used to find out the frequent 2-items set. From where
again support count of each item-set is calculated and only
those items whose support count is more than minimum
support count are kept and others are discarded. Next we find
out the frequent 3-item set and then frequent 4-item set until no
more frequent item sets can be generated. The final frequent
item set which is generated and satisfies the minimum support
count is our final frequent pattern.
B. Advantages and Disadvantages
The advantage of Apriori algorithm is that it is a simple
algorithm and can be implemented easily. But it still has some
disadvantages also. The main disadvantage is that here the
entire database needs to be scanned at each step. Also in this
algorithm, a large number of candidate keys are generates. And
if the database is very large than scanning in each step not only
consumes a lot of time but the generation of a large number
75

IJRITCC | March 2016, Available @ http://www.ijritcc.org

_______________________________________________________________________________________

International Journal on Recent and Innovation Trends in Computing and Communication

Volume: 4 Issue: 3

ISSN: 2321-8169
75 - 78

_______________________________________________________________________________________
candidate keys consumes a lot of memory also, which can be
sometimes limited3. Therefore this algorithm can work well for
small database but not for large database.
III.

IMPROVEMENTS OF APRIORI ALGORITHM

A number of improvements for Apriori algorithm have been

proposed to overcome the limitations of the algorithm. In this
section we will discuss some of the improvements and compare
them. We will take into consideration those improvements
which will reduce the number of scans and also the number of
candidate key generation. The minimum support is taken as 3
for the database given below.

I1I4

I4

T5,T7,T8

I2I3

I2

T2,T3,T4,T5,T6,T7,T8

Deleted

I2I4

I4

T5,T7,T8

I3I4

I4

T5,T7,T8

TABLE IV. FREQUENT 3-ITEM SET

Items

Support

I1I2I3

Item with
Min_support
I1

I1I3I4

I2I3I4

Transaction_IDs
T1,T3,T7,T9,T10

Deleted

I4

T5,T7,T8

Deleted

I4

T5,T7,T8

TABLE I. ORIGINAL DATABASE

Transaction

The number of scans for frequent 3-itemset = (5+3+3) = 11,

obtained from TABLE IV.

Items

T1

I1,I3,I7

T2

I2,I3,I7

T3

I1,I2,I3

T4

I2,I3

T5

I2,I3,I4,I5

T6

I2,I3

T7

I1,I2,I3,I4,I6

T8

I2,I3,I4,I6

T9

I1

T10

I1,I3

A. Algorithm for reducing the number of scan (Algorithm 1)

In this algorithm [4], we improve Apriori algorithm by
reducing the number of scans. In this algorithm the first step is
same as the classical Apriori algorithm. But in the second step,
from each of the frequent 2-item set we first find out the one
with minimum support count and then the transactions where
that item is present. Next only from that transaction we check
the frequent 2-items set and find the support count of individual
set. For all the next frequent item set we do the same. In this we
see the number of scans gets reduced. The algorithm works as
shown in TABLE II, TABLE III and TABLE IV.

B. Algorithm for reducing database size and number of scans

(Algorithm 2)
In this algorithm [5], Apriori algorithm was improved by
both reducing the number of scans as well as cutting down the
size of the database and removing some transactions which are
not required. As a result the search time gets reduced by two
times. In this algorithm also the first step is same as the original
Apriori algorithm. And the for frequent 2-item set, we first
delete all the transactions from the database which has less than
2 items, then for each individual 2-item sets we first find the
item with minimum support among the two then search for the
2-ietms sets only in those transactions where the minimum
support items are present and calculate their support count. We
remove those item-sets which support count less than minimum
support count. From the resultant set, frequent 3-item set s are
obtained and so on. The algorithm is illustrated with the help of
dataset provided in TABLE I. The improvements are shown in
the tables TABLE II, TABLE V, TABLE VI, TABLE VII
and TABLE VIII.
TABLE V. REDUCED DATABASE FOR FREQUENT 2-ITEM SET
Transaction

TABLE II. FREQUENT 1-ITEM SET

Items

Support

Transaction_IDs

I1

T1,T3,T7,T9,T10

I2

T2,T3,T4,T5,T6,T7,T8

I3

T1,T2,T3,T4,T5,T6,T7,T8.T10

I4

T5,T7,T8

I5

T5

Deleted

I6

T7,T8

Deleted

I7

T1,T2

Deleted

From TABLE II, it is found that the number of scans for

frequent 2itemsets = (5+5+3+7+3+3) =26.
TABLE III. FREQUENT 2-ITEM SET
Items

Support

I1I2

I1I3

Item with
Min_support
I1
I1

Transaction_IDs
T1,T3,T7,T9,T10
T1,T3,T7,T9,T10

Deleted

Items

T1

I1,I3,I7

T2

I2,I3,I7

T3

I1,I2,I3

T4

I2,I3

T5

I2,I3,I4,I5

T6

I2,I3

T7

I1,I2,I3,I4,I6

T8
T9

I2,I3,I4,I6

T10

I1,I3

I1

TABLE VI. FREQUENT 2-ITEM SET

Items

Support

I1I2

Item with
Min_support
I1

I1I3

I1

T1,T3,T7, T10

I1I4

I4

T5,T7,T8

I2I3

I2

T2,T3,T4,T5,T6,T7,T8

Transaction_IDs
T1,T3,T7, T10

Deleted

Deleted

76
IJRITCC | March 2016, Available @ http://www.ijritcc.org

_______________________________________________________________________________________

International Journal on Recent and Innovation Trends in Computing and Communication

Volume: 4 Issue: 3

ISSN: 2321-8169
75 - 78

_______________________________________________________________________________________
I2I4

I4

T5,T7,T8

I3I4

I4

T5,T7,T8

The number of scans for frequent 2itemsets = (4+4+3+7+3+3)

= 24.
TABLE VII. REDUCED DATABASE FOR FREQUENT 3-ITEM SET
Transaction

Items

T1

I1,I3,I7

T2

I2,I3,I7

T3

I1,I2,I3

T4

I2,I3

T5

I2,I3,I4,I5

T6

I2,I3

T7

I1,I2,I3,I4,I6

T8
T9

I2,I3,I4,I6

T10

I1,I3

Figure 1. Frequent 1-itemsets.

I1
Figure 2. Frequent 1-Itemsets after deleting the rows and columns.

The number of scans for frequent 1-itemset = 10x7=70

TABLE VIII. FREQUENT 3-ITEM SET

Items

Support

I1I2I3

Item with
Min_support
I1

T1,T3,T7

Deleted

I1I3I4

I4

T5,T7,T8

Deleted

I2I3I4

I4

T5,T7,T8

Transaction_IDs

The number of scans for frequent 3-itemset = (3+3+3) = 9.

C. Transaction Reduction and Matrix Method (Algorithm 3)
In this proposed algorithm [6], items (In) and transactions
(Tm) from the database are mapped into a matrix with size mxn.
In this matrix, transactions are represented by the rows and the
items are represented by the columns. The elements on this
matrix are either 0 or 1 and is decides as follows:
Martix = [aij] = 1, if in transaction i there is item j, and
Matrix = [aij] = 0, otherwise.
In the matrix, the sum of the row vector gives the sum of the
transactions (S-O-T) and sum of the column vector gives the
support count of each of the items. Now according to the
algorithm, we first generate the items sets by the above 2 rules.
Then for each frequent 1-item sets, we calculate the column
vector and check if or not the value is more than the minimum
support. If it is less, then the particular column is deleted. Then
as it is frequent 1-item set we delete all rows whose row sum is
equal to 1 or less than 1. Now from the resultant matrix we find
out the frequent 2-item set by joining and deleting columns
with column sum less than minimum support and row sum less
than or equal to 2. Similarly we proceed for all the other
frequent k-item sets by deleting columns with column sum less
than minimum support and row sum less than or equal to k,
until we find the frequent patterns. In this method the number
of scans is reduced and also we dont have to check the whole
database and has also reduced the I/O time spending in
scanning database but still it some overhead as it has to
maintain the updated database after each matrix generation.
The algorithm is explained in Figure 1, Figure 2, Figure 3,
Figure 4, Figure 5 and Figure 6.

Figure 3. Frequent 2-itemsets.

Figure 4. Frequent 2-Itemsets after deleting the rows and columns.

The number of scans for frequent 3-itemset = 9x6=54.

Figure 5. Frequent 3-itemsets.

Figure 6. Frequent 3-itemsets after deleting the rows and columns.

77
IJRITCC | March 2016, Available @ http://www.ijritcc.org

_______________________________________________________________________________________

International Journal on Recent and Innovation Trends in Computing and Communication

Volume: 4 Issue: 3

ISSN: 2321-8169
75 - 78

_______________________________________________________________________________________
TABLE XI. COMPARISION AMONG THE NUMBER OF SCANS

The number of scans for frequent 3-itemset = 4x3=12.

C. Algorithm for reducing number of scans and candidate
key formation (Algorithm 4)
In this algorithm [7], we improve Apriori algorithm by
reducing the number of scans and also the number of candidate
keys that are formed at each step. Here the steps for finding out
the frequent 1-item set and frequent 2-item set are same as the
classical Apriori algorithm. The algorithm is improved from the
next step where we find frequent 3-item set, frequent 4-item set
and so on. For instance, in the example shown below the
frequent 3-item sets that are formed by joining the frequent 2item sets are {I1,I2,I3}, {I1,I3,I4} and {I2,I3,I4}. But we will
not consider {I1,I2,I3} and {I1,I3,I4} as frequent 3-item sets
because the individual patterns are not frequent, that is {I1,I2}
and {I1,I4} are not frequent in each of the frequent 3-item set
and {I2,I3,I4} is frequent 3-item set because each of the
individual pair {I2,I3},{I3,I4} and {I2,I4} are frequent 2-item
set. TABLE II, TABLE IX and TABLE X. Though the number
of scans and candidate key generation is reduced in this
improved version but still efficiency of the algorithm has not
been improved at a high level.
TABLE IX. FREQUENT 2-ITEM SET
Item set

Support Count

I1I2

I1I3

I1I4

I2I3

I2I4

I3I4

Deleted

Deleted

Algorithm
Normal Apriori
algorithm
Algorithm 1
Algorithm 2
Algorithm 3
Algorithm 4

Number of scans
1-Item set

2-Item set

3-Item set

Total

70

60

30

160

70
70
70
70

26
24
54
60

11
9
12
10

107
103
136
140

TABLE XII. COMPARISION AMONG THE NUMBER OF

CANDIDATE KEYS GENERATED
Number of candidate keys
Algorithm
1-Item set

2-Item set

3-Item set

Total

Normal Apriori
algorithm

16

Algorithm 4

14

V. CONCLUSIONS
The Apriori algorithm is the most basic algorithm that was
developed for finding out the frequent patterns in large
databases which suffered from certain limitations. In this paper,
four proposals for the improvement of Apriori algorithm were
discussed that have successfully overcome this limitations and
their comparisons have been done and shown. But still these
algorithms needs to be optimised more in terms of time
consumption, memory requirement, efficiency and reduction in
terms of number of scans.
REFERENCES
[1]

The number of candidate keys generated = 6.

[2]

TABLE X. FREQUENT 3-ITEM SET

Item set

Support Count

I2I3I4

[3]

[4]

The number of candidate keys generated = 1.

IV. RESULTS AND DISCUSSION
[5]

In the above section we have discussed few algorithms

developed to overcome the limitations of the Apriori algorithm.
The result of all the algorithms are same, that is {I2, I3, I4}.
And all the above algorithms have successfully reduced the
number of scans. Some of them have also reduced the size of
the database and candidate keys that are generated. In TABLE
XI, we will compare the number of scans required in all of
these algorithms. Hence we can conclude that the second
algorithm has been able to reduce the number of scans to a
huge extent. In TABLE XII, we compare the number of
candidate keys generated. In this paper the comparison was
made on a relatively small database. It can be seen that
difference will be significant in the reduced values of the
number of scans and the number of candidate keys generated
on a larger database.

[6]

[7]

J. Han, M. Kamber, J. Pei Data Mining Concepts and

Techniques, Morgan Kaufmann Publisher, Third Edition, Year
2012.
S. Nasreen, M. A. Azamb, K. Shehzada, U. N. M. Ali
Ghazanfara, Frequent Pattern Mining Algorithms for Finding
Associated Frequent Patterns for Data Streams: A Survey
Procedia Computer Science 37, pages 109 116, Year 2014.
R. Agrawal and R. Srikant. Fast algorithms for mining
association rules, In Proc. 1994 Int. Conf. Very Large Data
Bases, pages 487499, Santiago, Chile, September 1994.
M. Al-Maolegi1, B. Arkok, AN IMPROVED APRIORI
ALGORITHM FOR ASSOCIATION RULES, International
Journal on Natural Language Computing (IJNLC), vol. 3, No.1,
February 2014.
S. Aggarwal and R. Sindhu, An approach to improve the
efficiency of apriori algorithm. PeerJ PrePrints 3:e1410, 2015
V. Mangla, C. Sarda, S. Madra, Improving the efficiency of
Apriori Algorithm in Data Mining, International Journal of
Engineering and Innovative Technology (IJEIT), Vol. 3, Issue 3,
September 2013.
X. Fang, An Imroved Apriori Algorithm on the Frequent
Itemsets, International Conference on Education Technology
and Information System, Sanya, China, 845 848, 2013.

78
IJRITCC | March 2016, Available @ http://www.ijritcc.org

_______________________________________________________________________________________