Multiobjective feature selection for key quality characteristic identification in production processes using a nondominated-sorting-based whale optimization algorithm

Abstract

Identifying key quality characteristics (QCs) in production processes is essential for product quality control and improvement. This paper proposes a multiobjective wrapper-based feature selection (FS) method for key QC (KQC) identification on unbalanced production data using a novel modified nondominated-sorting-based whale optimization algorithm (MNSWOA) and the ideal point method (IPM). In the proposed approach, the FS problem is defined as maximizing the geometric mean (GM) measure and minimizing the feature (QC) subset size. To solve the defined FS problem, MNSWOA is adopted first to find a set of candidate solutions (feature subsets), and then IPM is adopted to select the final solution. In MNSWOA, a modified fast nondominated sorting approach is proposed to adapt the single objective whale optimization algorithm to the multiobjective scenario. Moreover, a uniform reference solution selection strategy and the mutation operations are embedded in MNSWOA to improve its search performance. Experimental results on four unbalanced production datasets show that the proposed FS method performs effectively and efficiently for KQC identification. Further comparisons show that MNSWOA obtains better search performance than benchmark multiobjective optimization methods, including a modified NSGA-II, SPEA2, MOEA/D, NSPSO and CMDPSO.