CONSTRAINT FREQUENT PATTERN MINING

Abstract

As our world is in its information era, a huge amount of data is accumulated everyday. The real universal challenge is to obtain actionable knowledge from a large amount of data. Data mining is an emerging research direction to meet this challenge. Mining of frequent itemsets is a fundamental data mining task. This area of study mainly adopts an Apriori-like candidate set generation-and-test or Frequent Pattern growth like without candidate set generation approach. Earlier researches have proposed many efficient algorithms for mining of frequent patterns. Recent studies highlight the importance of using constraints to focus the mining process to mine only relevant itemsets. This dissertation work addresses the problem of mining frequent patterns efficiently and effectively. First, this includes concept of frequent pattern mining by implementation of various frequent pattern mining algorithms, followed by comparative analysis, based on values of various parameters like time required to find itemsets and number of database scans, obtained from experimental results. Then, the concept of constraint frequent pattern mining by considering item and size constraint is elaborated and implemented, along with the discussion of various categories of constraints like monotonic and convertible, based on their properties. Finally, based on the results obtained from the experiments, comparative analysis of the various approaches of constraint frequent mining algorithms is performed. This dissertation work also proposes a combined approach of data filtering (size constraint) with item constraint for finding itemsets with multiple constraints. All the algorithms are implemented in JAVA language and tested on windows XP platform. iii