INTERPRETABILITY AND MEAN-SQUARE ERROR PERFORMANCE OF FUZZY INFERENCE SYSTEMS FOR DATA MINING

摘要

Over the years, many methods have become available for designing fuzzy inference systems from data. Their efficiency is usually characterized by a numerical index, the mean-square error. However, for human-computer cooperation, another criterion is needed; the rule of interpretability. This paper analyses two kinds of fuzzy inference system: fuzzy clustering algorithms to organize and categorize data in homogeneous groups, and grid partitioning (generated from data or given by experts) of the multidimensional space. The methods are compared according to mean-square error performance and an interpretability criterion. Simulation results carried out on a forecasting problem associated with stock market are included.