A modified NEFCLASS classifier with enhanced accuracy-interpretability trade-off for datasets with skewed feature values

The accuracy-transparency trade-off is one of the most notable challenges when applying machine learning tools in the medical domain. NEFCLASS is a popular NEURO-FUZZY classifier in medical diagnosis systems. NEFCLASS performs increasingly poorly as the data skewness increases. This paper presents a combined approach to improve the classification accuracy and interpretability of the NEFCLASS classifier, when feature values of the training and testing datasets exhibit positive skewness. The proposed model consists of two steps. Firstly, a modified NEFCLASS classifier embedded with a choice of two alternative discretization methods, MME and CAIM is implemented. Secondly, we devised a new rule pruning method based on the Habermans' adjusted residual to reduce the size of the resulting ruleset. This rule-pruning method improves the interpretability of NEFCLASS without significant accuracy deterioration. Moreover, a hybrid approach combining the two approaches provides a considerable improvement in classification accuracy and transparency of NEFCLASS on skewed data. (C) 2020 Elsevier B.V. All rights reserved.

Automated summary

The paper introduces a combined approach to improve the classification accuracy and interpretability of the NEFCLASS classifier in the presence of skewed data by modifying the classifier and implementing a new rule-pruning method.