Feature Selection Based on Clustering Valid Analysis with Fuzzy-Rough Set

Fuzzy c-means clustering is introduced to fuzzify the continuous attributes of fault features in an attempt to decline information loss during the course of discretization. Clustering valid analysis is utilized to obtain the optimal number of clusters, and by this way, the shortcoming of current approaches that number of clusters need to be determined artificially is overcome. Experiments of fault diagnosis on aero-engines show that the proposed approach of fault feature selection is feasible.