Feature Selection Algorithm Used to Classify Faults in Turbine Bearings

Feature Selection is a very important step that select a few number of feature used for the classification in order to reduce execution time, to improve accuracy and to enhance performance of the identification system. In this paper we propose new feature selection methods by combining between relief, mutual information and sequential selection. The new approach is compared with other existing and we demonstrate some improvement when they are applied to a random dataset and on real data acquired from wind turbine bearings aiming to detect fault in the turbine using vibration signal.