Rule-based Explainable Knowledge Acquisition by Multiobjective Evolutionary Machine Learning

2022 Fiscal Year Final Research Report

• Project/Area Number: 19K12159
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 61040:Soft computing-related
• Research Institution: Osaka Metropolitan University (2022); Osaka Prefecture University (2019-2021)
• Principal Investigator: Nojima Yusuke 大阪公立大学, 大学院情報学研究科, 教授 (10382235)
• Project Period (FY): 2019-04-01 – 2023-03-31
• Keywords: 知識獲得 / 進化型機械学習 / 解釈可能性

Outline of Final Research Achievements

In this study, we developed fuzzy rule-based classifier design methods considering accuracy, interpretability, reliability, and fairness. There is a trade-off between accuracy and interpretability. Accurate knowledge is difficult to interpret, while knowledge that is easy to interpret has low accuracy. In this study, we developed a method for improving the accuracy of fuzzy classifiers with high interpretability. We also developed a fuzzy classifier with a reject option that does not output low-confidence discriminant results. We also studied fuzzy classifier design that considers the fairness of the knowledge obtained from the data. The extensions to class imbalance data and multi-label data are discussed. The effectiveness of the above-developed methods is demonstrated through computational experiments.

Free Research Field

計算知能

Academic Significance and Societal Importance of the Research Achievements

現在の識別器設計に関する研究は，深層学習やアンサンブル識別器が主流であり，識別結果の解釈可能性の低さを改善する研究が数多く行われいる．一方，本研究は言語的解釈可能な識別器設計手法の高精度化であり，他の機械学習研究者や利用者に異なるアプローチを提供可能である．また，機械学習分野において近年注目されている精度，解釈可能性，信頼性，公平性という観点を，本研究で網羅的に取り扱っており，実データ解析手法として社会実装や，他の機械学習手法の改良や新たな手法の開発への一助となることが期待できる．