A novel switching function approach for data mining classification problems

Abstract

Rule induction (RI) is one of the known classification approaches in data mining. RI extracts hidden patterns from instances in terms of rules. This paper proposes a logic-based rule induction (LBRI) classifier based on a switching function approach. LBRI generates binary rules by using a novel minimization function, which depends on simple and powerful bitwise operations. Initially, LBRI generates instance codes by encoding the dataset with standard binary code and then generates prime cubes (PC) for all classes from the instance codes by the proposed reduced offset method. Finally, LBRI selects the most effective PC of the current classes and adds them into the binary rule set that belongs to the current class. Each binary rule represents an If–Then rule for the rule induction classifiers. The proposed LBRI classifier is based on basic logic functions. It is a simple and effective method, and it can be used by intelligent systems to solve real-life classification/prediction problems in areas such as health care, online/financial banking, image/voice recognition, and bioinformatics. The performance of the proposed algorithm is compared to six rule induction algorithms; decision table, Ripper, C4.5, REPTree, OneR, and ICRM by using nineteen different datasets. The experimental results show that the proposed algorithm yields better classification accuracy than the other rule induction algorithms on ten out of nineteen datasets.