Enhanced Prediction of Manufacturing Quality Ratings Using Optimized Stacking Ensemble Modeling

Abstract

The predictive modeling of quality in manufacturing is vital in improving efficiency, cutting costs, and attaining zero-defect production. This paper addresses this requirement through an empirical comparison of machine learning models to predict manufacturing quality ratings. On a simulated dataset of 3,957 samples with five important features of the process, this study compared the performance of six basic models (AdaBoost, Bagging, Decision Tree, Gradient Boosting, K-Nearest Neighbors, and XGBoost) and two more advanced ensemble models (Averaging and Stacking). The findings demonstrated that the Stacking Ensemble model outperformed all other candidates, with higher performance in terms of an R2 of 0.999 and the lowest values of error (MSE: 0.004, RMSE: 0.065, MAE: 0.016) in the test set. Moreover, SHAP analysis of the Stacking ensemble as the best model revealed that Material Transformation Metric (MTM) and Temperature (T) were the top features having a significant impact on the quality of products. The analysis finds that the Stacking Ensemble method provides a valid and very effective framework for predicting quality, which is helpful in the optimization of the manufacturing process.