A GENETIC BASED RESEARCH FRAMEWORKTO DISCOVER OPTIMAL FREQUENT PATTERNS USING ASSOCIATION RULE MINING

Abstract

The rapid advances in data generation, availability of automated tools in data collection and continued decline in data storage cost enabled with high volumes of data. In addition, the data is non scalable, high dimensional, heterogeneous and complex in its nature. This situation creates inevitably increasing challenges in extracting desired information. Thus, Data mining evolves into a fertile area and got the focus by many researchers and business analysts. Data mining is a methodology the blends traditional techniques with sophisticated algorithms. Among all, the association rule mining is efficient pattern discovery technique, which finds hidden, valid, novel, useful, understandable, interesting and ultimately correlated patterns in large databases. Such correlated rules create great business value to any organization as they make use in decision making process. However, in real time applications the correlation changes continuously as the source data updates dynamically. This motivation necessitates finding and updating the frequent item sets with different supports efficiently and optimally. In order to overcome the challenges inherited in conventional association rule mining, the authors in the present paper propose an Optimal Frequent Patterns System (OFPS). The OFPS takes radically a different approach and design as a three-fold system that discovers optimal frequent patterns efficiently, using the genetic algorithm. Initially, the first-fold of OFPS focuses on preparation of domain specific data that includes data selection, cleaning, integration and transformation under the guidance of knowledge expert. Subsequently, the second-fold of OFPS emphasizes on construction of a Frequent Pattern Tree (FP-Tree) and then discovery of frequent patterns by exploring the tree in the bottom-up fashion to facilitate rapid access of individual frequent patterns quickly. The third-fold of OFPS finally concentrates on generation of optimal frequent patterns using genetic algorithm that simulates biological evaluation procedure having the self learning capability. To validate the performance of proposed OFPS in several orders of magnitude, many experiments were conducted and results have proven this as claimed.