Enhanced cAntMinerPB Algorithm for Induction of Classification Rules using Ant Colony Approach

Abstract

Mining classification rules from data is a key mission of data mining and is getting great attention in recent years. Rule induction is a method used in data mining where the desired output is a set of Rules or Statements that characterize the data. Within Rule Induction model, Swarm Intelligence (SI) is a technique where rules may be discovered through the study of joint behavior in decentralized, self-organized systems, such as ants. Ant-Miner is a rule induction algorithm that uses SI techniques to form rules. The main idea of this study is to discover the suitability of ant colony optimization for constructing accurate classifiers which can be learned in practical time even for big datasets. The cAntMinerPB algorithm is an extension of the cAntMiner algorithm. The main task is to modify the existing algorithm cAntMinerPB to allow each rule to dynamically select rule quality evaluation function and to improve the accuracy and preserving rule list simplicity. In this study, we examine the use of different rule quality evaluation functions for rule quality assessment prior to pheromone update and check how the use of different evaluation function affects the output model in terms of predictive accuracy and model size. In experimental results, we use 10 different rule quality evaluation functions on 12 benchmark datasets and found that predictive accuracy obtained by new proposed method is statistically significantly higher than the predictive accuracy of existing algorithm.