Bi-objective variable selection for key quality characteristics selection based on a modified NSGA-II and the ideal point method

A product, or even a part, may contain hundreds of quality characteristics (QCs), but not all of them are key characteristics that determine product quality. Selecting possible key quality characteristics (KQCs), while eliminating redundant or noisy QCs, is a necessary step before implementing quality control or improvement tools. In this paper, we propose a two-phase variable selection algorithm based on a modified non-dominated sorting genetic algorithm II (NSGA-II), a multi-objective evolutionary algorithm, and the ideal point method (IPM) for KQC selection. In modified NSGA-II, we use a modified fast non-dominated sorting approach to increase the diversity of population in the evolutionary process. The uniqueness of the algorithm is that IPM can select KQC sets with few QCs from the candidate QC subsets found by the modified NSGA-II. Experimental results show that the proposed algorithm outperforms benchmarked KQC selection algorithms in terms of classification accuracy rates and number of noisy or redundant QCs.