Hybridization of Fuzzy Sets and Machine Learning Models for Solving Regression and Classification Problems

This paper presents a novel approach to hybridizing fuzzy sets with machine learning models. The proposed hybrid system is not based on fuzzy rules, so the curse of dimensionality that appears in fuzzy rule-based systems is avoided. The system consists of a fuzzification module and a predictive model realized by a machine learning technique. The fuzzification module transforms each observational variable into fuzzy information described by the fuzzy basis functions. The predictive model uses these fuzzy data as inputs and computes an output using machine learning techniques. In the proposed solution, the fuzzification module can be combined with any predictive model, for example, a neural network, support vector machine, linear discriminant analysis, etc. In tuning the system, a feature selection method is used, which employs fusion of rankings based on a Borda voting system. The results of experiments confirm the efficacy of the presented method and show that the proposed system can improve the results obtained by non-fuzzy systems.