Predictive Modeling of Vicat Softening Point for Low-Density Polyethylene Using GBM, XGBoost and AdaBoost: A Comparative Analysis

This study provides a comparative analysis for predicting Vicat softening temperatures of Low-Density Polyethylene (LDPE), which is one of the versatile polymers used extensively across various industrial sectors. LDPE exhibits unique properties, such as flexibility, electrical insulating characteristics, and low melting point. The manufacturing process demands rigorous quality controls, involving extensive laboratory product testing, requiring significant time, labor, and cost investments. In this study, we explored the potential of machine-learning-based predictors to alleviate or reduce these challenges. Our analysis focused on the accuracy and processing time of predicting models based on three prominent boosting methods: Gradient Boosting Machines (GBM), Extreme Gradient Boosting (XGBoost), and Adaptive Boosting (AdaBoost). We collected the laboratory testing results from one of the largest polymer manufacturers in Southeast Asia: our data set comprised 71 features. Based on the comparison results, we concluded that XGBoost exhibits superior predictive performance (in terms of MAE, MSE, and RMSE) compared to GBM and AdaBoost, indicating its potential in time saving, labor, and cost reduction in the manufacturing process. Both XGBoost and AdaBoost incurred maximum errors below 2.9, aligning with the industry testing standard. Notably, Adaboost incurred slightly lower maximum errors in comparison to XGBoost. Furthermore, we presented the top 10 significant features highlighted by the XGBoost models.