Efficient approach for the design of transparent fuzzy rule-based classifiers

In the last few years a number of studies have proposed algorithms that can obtain fuzzy systems which are simple and easy to read, while maintaining quite a high level of accuracy. Following this philosophy, the paper presents a simple, new approach based on Genetic Algorithms, with the aim of selecting the features and tuning the parameters of a fuzzy classification algorithm. From the results obtained by the optimized classifier a transparent, efficient fuzzy system is generated using simple heuristic methods. The main features of the approach are accuracy, scalability, adaptability and expandability. Comparative examples based on three data sets well known in the pattern classification field are given, showing that the approach leads to classifiers with a small number of transparent, readable rules, which are less complex than those reported in the literature with comparable or better accuracy.