A Genetic Algorithm for Discovery of Association Rules

Abstract

A genetic algorithm is proposed in this article for discovery of association rules. The main characteristics of the algorithm are: (1) The individual is represented as a set of rules (2) The fitness function is a criteria combination to evaluate the rule's quality - high precision prediction, comprehensibility and interestingness -- (3) Subset Size-Oriented Common feature Crossover Operator (SSOCF) is used in the crossover stage (4) mutation is calculated through non-symmetric probability and selection strategy through tournament and (5) the algorithm was implemented using the library lambdaj. Finally, the genetic algorithm effectiveness and the quality of the rule in the experimental results are shown.