Fequent Pattern Recognization From Stream Data Using Compact Data Structure

Abstract

Mining frequent pattern from stream data is a challenging task. Finding frequent pattern
from data streams have been found to be useful in many application such as stock market prediction,
sensor data analysis, network traffic analysis, e-business and telecommunication data analysis.
Frequent Pattern Stream tree [1] is used for maintaining frequent pattern over a period of time using
modified FP tree algorithm. This approach maintains tilted time window at each node which
consumes larger space. Compact Pattern Stream Tree [2] assumes that only current patterns are of
importance and uses sliding window protocol for maintaining it. This approach does not give
importance to past frequent patterns.
The aim of this research paper is to combine the features of frequent pattern stream tree and
Compact pattern stream tree to find the frequent pattern over a period of time. The research assumes
the patterns that are infrequent for a period can become frequent in the future. It is proposed to
generate a Compact Frequent Pattern Stream tree that maintains tilted time window at the tail node
and at partition node that keep track of the pattern over a period of time. The pattern whose support
count is greater than predefined threshold is considered to be frequent. The proposed approach has
proved advantage of reduction in space utilization, compaction in space utilization and consideration
of past patterns that may become frequent. The system will generate frequent pattern with compact
data structure and in a time efficient manner.