Dominant Feature Selection Methods for Root Cause Identification in Injection Moulding Process

This paper introduces a hybrid feature selection and classification validation approach for anomaly detection to identify major failure root causes using supervised classification validation. The proposed method is tested on real industry datasets from the manufacturing injection moulding process. Results indicate that the SMOTE technique effectively generates a balanced dataset for the minority class in both testing and training subsets, thereby improving model accuracy without altering the raw data samples. Pearson's correlation emerged as the best feature selection method when compared to Decision Tree and Recursive Feature Elimination methods. Random Forest was identified as the optimal classification model for anomaly detection using the case study datasets. This hybrid feature selection approach demonstrates that the study results can be applied to identify major root causes, thereby improving the quality performance of the injection moulding process.