Mathematical and machine learning analysis of high-density polyethylene thermal degradation

Thermogravimetric analysis (TGA) is widely used to characterize polymer degradation, but repeated experiments and conventional single-step kinetic models can become inefficient or inadequate when degradation is nonlinear, multistage, atmosphere-dependent, or evaluated outside the conditions used for calibration. This study developed a machine learning (ML) regression framework to predict the percentage mass remaining during high-density polyethylene (HDPE) degradation. Twenty-five TGA specimens (8–20 mg) yielded 237,154 observations; time, temperature, and initial mass were used as predictors. Random Forest, Extra Trees Regressor (ETR), Support Vector Regressor (SVR), eXtreme Gradient Boosting (XGBoost), Adaptive Boosting (AdaBoost), Categorical Boosting (CatBoost), Natural Gradient Boosting (NGBoost), and a multilayer perceptron artificial neural network (MLP–ANN) were compared with a hybrid NGBoost–ETR model. Performance was assessed using the coefficient of determination (R²), mean squared error (MSE), root mean squared error (RMSE), and mean absolute error (MAE). The hybrid model achieved R² = 0.999987, MSE = 0.027042, RMSE = 0.164446, and MAE = 0.015879 on the dataset. Shapley Additive Explanations (SHAP) ranked time and temperature as the dominant predictors, while initial mass had a smaller contribution. The optimized point-prediction model was deployed through a Django web interface. The framework provides rapid within-domain interpolation of HDPE TGA profiles; external validation across polymer grades, heating programs, and laboratories remains necessary before replacing experimental TGA.