Multiple Disjunctions Rule Genetic Algorithm (MDRGA): Inferring Non-Linear IF-THEN Rules in Non-Linear Datasets

Abstract

Genetic Algorithms (GAs) are computational search methods based on Darwin's evolutionary theory. In the present study, the MDRGA (Multiple Disjunctions Rule Genetic Algorithm) is proposed as a tool to induce non-linear IF-THEN classification rules from non-linear datasets, which can be used as a classification system. The main goal of MDRGA is to allow the discovery of concise, yet accurate, non-linear high-level IF-THEN rules balancing prediction precision, comprehensibility and interpretability. The results show that the MDRGA is promising and capable of extracting useful high-level knowledge with good precision values. The classification accuracy of proposed MDRGA was compared with other GA-based methods (CEE and NLCEE) and traditional classification methods (J48, Random Forest, PART, Naive Bayes and IBK) in four non-linear datasets (Sonar, Diabetes, Iris and Breast-W) downloaded from UCI Machine Learning Repository and the MDRGA obtained the best classification accuracy results for all datasets.