Data mining with a parallel rule induction system based on gene expression programming

摘要

A parallel rule induction system based on gene expression programming (GEP) is reported in this paper. The system was developed for data classification. The parallel processing environment was implemented on a cluster using a message-passing interface. A master-slave GEP was implemented according to the Michigan approach for representing a solution for a classification problem. A multiple master-slave system (islands) was implemented in order to observe the co-evolution effect. Experiments were done with ten datasets, and algorithms were systematically compared with C4.5. Results were analysed from the point of view of a multi-objective problem, taking into account both predictive accuracy and comprehensibility of induced rules. Overall results indicate that the proposed system achieves better predictive accuracy with shorter rules, when compared with C4.5.