Fusion of Feature Selection Methods for Improving Model Accuracy in the Milling Process Data Classification Problem

The current study addresses the problem of feature selection performed for the data set collected in the milling process. The data consists of 1709 records with 44 statistical parameters computed on the basis of the measured input signals from the accelerometer mounted on a lower bearing of the spindle of Haas VM-3 CNC machining centre and the acoustic emission sensor mounted in the machine cabin. A new feature selection approach is proposed which is based on the fusion of three filter methods: Pearson’s linear correlation coefficient, ReliefF and single decision tree. By means of the introduced combined ranking set, the most significant features are stored and then employed to create the reduced data set. The validity of the proposed solution is tested by computational intelligence models in original and reduced data classification task. Based on the experimental study the efficacy of the approach is confirmed.