Performance Comparison of Evaluation Measures in Michigan-style Fuzzy Genetics-based Machine Learning for Imbalance Datasets

Abstract

Fuzzy genetics-based machine learning generates fuzzy classifiers with high interpretability. However, there is a problem that the classification performance for the minority class is not high because the evaluation measure in the original algorithm does not consider the class distribution of the dataset. This study examines the effects of various evaluation measures on the classification performance for the minority class in the Michigan-style fuzzy genetics-based machine learning. We also compare between the Michigan-style fuzzy genetics-based machine learning algorithm considering the minority class and other classifier design algorithms with an oversampling method.