Extraction of Critical-to-quality Characteristics of Complex Products using mRMR and KPCA

In the production of intricate products, the extraction of critical-to-quality characteristics (CTQs) is crucial for achieving effective quality control and defect prevention. The presence of numerous irrelevant or redundant quality characteristics in the manufacturing process can obscure critical information, thereby diminishing the effectiveness of monitoring and prediction. In response to this challenge, the present work introduces a CTQ extraction approach that combines Minimum Redundancy Maximum Relevance (mRMR) for feature selection with Kernel Principal Component Analysis (KPCA) for nonlinear dimensionality reduction. Firstly, the mRMR algorithm is employed to select representative CTQs from high-dimensional quality characteristics through optimizing feature-target relevance and simultaneously constraining redundancy among features. Secondly, KPCA is applied to extract the selected characteristics to better capture the latent nonlinear structures within the data. Finally, the SECOM dataset, consisting of semiconductor products, is utilized to evaluate the efficacy of the proposed approach using the XGBoost model, with results showing that mRMR effectively selects representative CTQs, KPCA captures nonlinear relationships among features, and XGBoost achieves superior predictive performance while maintaining reduced dimensionality.