Entropy-Driven Data Reduction for Nonlinear System Identification Using ANFIS

Intelligent systems theory embraces uncertainty as an intrinsic feature of information and tries to address problems adaptively without forcing them into some predefined formula. Instead, the theory examines systems based on their inherent characteristics, uncovering knowledge that is usually represented in data sets. While volume grows, processing time increases rapidly; the challenge is in how modern methods detect redundancy in the datasets and arrive at a reduced yet effective subset of data. This paper deals with the aforementioned challenge. We propose a methodology that combines knowledge granulation, defined using Yager’s fuzzy entropy, with the ANFIS framework to validate the effectiveness of the reduced data subset in maintaining high performance with reduced computational overhead in system identification through the core input-output relationships of the dataset. The proposed method enables faster and more efficient knowledge extraction by alleviating redundancy compared to traditional techniques. As for demonstration of the effectiveness of it, the ANFIS methodology is applied to the so-called family of commonly met arbitrary function estimation problems. Through simulation results, this new approach confirms the effective preservation of the approximation capability on identifying nonlinear systems while seriously relaxing computational costs.