Discovery of Novel Association Rules Based on Genetic Algorithms

Abstract

Association rule mining is a data mining task that attempts to discover interesting knowledge from huge databases. Data mining researchers have studied subjective measures of interestingness to reduce the volume of discovered rules to ultimately improve the overall efficiency of KDD process. Genetic algorithm (GA) based on evolution principles have found its strong base in mining association rules (ARs). In this paper, confidence and novelty measures have been pushed into a genetic algorithm in order to generate association rules form huge data and discover a novel and hence interesting knowledge to support decision makers. A hybrid approach that uses objective and subjective measures has been used in this paper to quantify novelty of association rules during generation process in terms of their confidence and deviations from the known rules. The proposed approach has a flexible chromosome encoding involve Apriori algorithm where each chromosome should be compute its support and confidence values to performs prune process of week chromosomes. In addition each chromosome differs from another in terms of number of items and classes. The proposed approach has been experimented using real-life public datasets and tested using real life applications. The experimental results have been presented and quite promising.