SUPPLIER PERFORMANCE EVALUATION PREDICTIVE MODEL FOR DIRECT MATERIAL USING MACHINE LEARNING APPROACH IN SEMICONDUCTOR MANUFACTURING

Abstract

In semiconductor manufacturing, evaluating supplier performance for direct materials is often unreliable and biased, failing to accurately represent suppliers' true performance. The objective of this paper is to present a data-driven Supplier Performance Evaluation (SPE) predictive model for direct material in semiconductor manufacturing. By using multiple machine learning techniques, the model provides unbiased evaluations of supplier performance. The model uses six machine learning methods: Logistic Regression, Support Vector Machine, Naïve Bayes, Generalized Linear Model, Decision Tree, and Random Forest. . The results show that Logistic Regression outperforms the other techniques with regards to analyzing both data from incoming material checks and the assembly in-process. The AUC-ROC value is 0.993 from Logistic Regression, proving that the model can identify material withdrawal trends effectively. In conclusion, the resulting model can enhance monitoring, risk management, and proactive supplier management, which leads to an efficient supply chain.