Abstract: uses vertical dataset and Depth first search

Abstract:

Frequent pattern mining used to create a frequently
used item sets which requires a lot of complexity and memory. In real world
many applications uses frequent patterns. Mining frequent patterns with the use
of bitmap an algorithm called bitmap .This paper proposes an algorithm which is
an improved form of FBSB algorithm to mine the frequent patterns.It uses bitmap
for storing of position of item in a sequence which is used to create next
sequence item sets. This algorithm presents (k+1) candidate sequences which are
not adjacent to k sequences.in the experiments this algorithm can achieve a
better performance than FBSB algorithm.

Introduction

Since the introduction of frequent pattern mining many
algorithms were being proposed for performing the task. These algorithms can be
classified in to two categories: algorithms using the horizontal dataset format
such as Apriori and FP-growth and algorithms using vertical dataset such as
Eclat. Mining frequent patterns is an necessary component in data mining tasks.
Apriori is used for discovering all significant item sets in a large database
of transactions, but there should be repeated scanning of database which
increases run time. FP-growth, for mining the complete set of frequent patterns
by pattern fragment growth uses a tree structure and will be a divide and
conquer method will be used to reduce the mining task. Another classic
algorithm is Eclat. It uses vertical dataset and Depth first search which
reduces memory requirements but too expensive. DBV Miner algorithm which uses
look up table, support can be computed by the intersection of two DBVs. Mining
frequent patterns uses a minimum support threshold. However setting this
threshold is too large or small, it influences the number of patterns generated

The FBSB+ algorithm presented here is to mine frequent
item sets with high complexity and less time. It first forms a bitmap based on
the item sets and provides the end position of an item. Then the support of
each item is calculated. Then frequent 2-sequences are generated using bitmap in
which the position of the item joined to the item in frequent 2- sequence
should be lower. Then the bitmap for frequent -2 sequences is created in which
the end position of the sequence is placed in bitmap. There is no repetition of
scanning database and frequent item sets can be generated. Experimental results
show that FBSB+ is better when compared with the FBSB.