A Hybrid Feature Selection Approach Based on ReliefF-FC-SS Algorithm for Multi-feature Data

Nowadays, the extensive application of PHM technology makes mechanical equipment more precise and automated, but the massive data generated during the operation also significantly increase the difficulty of evaluating equipment operation status. In order to accurately and efficiently identify the categories of equipment operating states, it is necessary to extract a high-quality feature subset from the original high- dimensional feature space. Therefore, this paper improves the traditional ReliefF algorithm by introducing feature correlation and stepwise selection, and proposes ReliefF-FC-SS algorithm. To verily the performance of ReliefF-FC-SS algorithm, this paper applies three classical feature selection approaches and the approach based on ReliefF-FC-SS algorithm to nine public datasets. The experimental results show that the proposed approach can capture more information with fewer features on the premise of ensuring classification accuracy.