Weighted Feature Selection Method for Improving Decisions in Milling Process Diagnosis

In this article, a new feature selection method is introduced. It is based on the weighted combined ranking score which fuses feature significance provided by three approaches: the Pearson’s linear correlation coefficient, ReliefF and single decision tree. During the successive steps, we eliminate the least significant features using binary weights corresponding to individual features. The utilized data set is represented by 1709 records and 44 attributes determined based on the signals acquired in the milling process. The efficiency of the proposed method is tested on reduced and original data set by a multilayer perceptron classifier. The obtained results confirm the usefulness of the solution.