Different Support Pattern Selection Methods for Heuristic Rule Generation from Multiple Patterns in Michigan-style Fuzzy Genetics-based Machine Learning

Abstract

In heuristic rule generation of our Michigan-style fuzzy genetics-based machine learning (GBML), each rule has been generated by using a single pattern. Recently, we proposed a modified heuristic rule generation from multiple patterns. We demonstrated the advantage of the use of multiple patterns over the standard single pattern-based approach. However, our previous approach can be further improved by utilizing available information such as the class of each pattern and the distance between each pair of patterns. In this paper, we examine the effects of different rule selection methods considering the pattern class and distance on the generalization ability of our Michigan-style fuzzy GBML.