Methods for Binarization of Metaheuristic Algorithms for Constructing Compact Fuzzy Classifiers of Medical Data

Constructing interpretable machine learning models that are able to explain the result of the inference process is an important factor to increase the level of confidence and trust in the machine learning models and their results, especially in medicine. Fuzzy classifiers are interpretable in nature because they consist of IF-THEN rules and use fuzzy terms used in colloquial speech. For the construction of fuzzy classifiers, the authors use a three-stage construction method consisting of a structure generation stage, a feature selection stage, and a rule parameter optimization stage. This paper focuses on the feature selection stage. At this stage, swallow swarm and leaping frogs binary metaheuristic algorithms are used for feature selection by wrapper method. Various implementations of binarization algorithms are proposed to convert the search space from continuous to discrete. Experiments on classifier construction and statistical evaluation of the effectiveness of the resulting models were performed, which showed a significant advantage in feature reduction compared to genetic fuzzy systems.