Study on interpretable fuzzy classification system based on neural networks

This paper describes a comprehensive method to construct fuzzy classification system considering both precision and interpretability. Fuzzy classification system, initialized by modified Gath-Geva fuzzy clustering algorithm, is transformed into neural network. After training the neural network, fuzzy sets similarity measure is adopt to merge redundant fuzzy sets to improve interpretability, and a constraint genetic algorithm is applied to improve precision. The simulation result on Iris data problem demonstrates the effectiveness of the proposed method.