A Comparative Investigation Introducing Regularization Techniques in Linear Regression Models for Quality Prediction in Forming Technology

In this investigation, linear models used for quality prediction of a final product are compared and evaluated using data from a real manufacturing process in forming technology (i.e., flexible rolling process). Two alternative methods for simplifying the feature selection method for the quality prediction model of manufactured blanks are presented. This work proposes implementing L1 and $\mathbf{L} 2$ regularization techniques in the original regression model. The method is then evaluated based on model complexity and performance metrics using the final predictions. By comparing these indicators, the effectiveness and benefits of the proposed method are confirmed. A simplification in the model-building effort and feature selection process is developed while providing an efficient and comparable accuracy in the predicted quality of the manufactured blanks.