Data-Driven Hybrid Approach Using Hyperparameter-Optimized Ensemble and Explainable Machine Learning for Assessing Pyrolysis Efficiency of Waste Tires

Predicting pyrolysis oil yield from waste tires is a complex challenge due to the nonlinear interactions between feedstock composition and process parameters. Therefore, this study suggests the use of Decision Tree, Linear Regression, and XGBoost models to create an interpretable machine learning framework to estimate pyrolysis oil yield based on key features such as pyrolysis temperature, hydrogen, oxygen, nitrogen, volatile matter concentrations, and ash content. As a result, XGBoost outperformed the other models, with R 2 values of 0.965 (training) and 0.914 (testing), low root mean squared errors, and low mean absolute percentage errors. Furthermore, the Shapley Additive ExPlanations study showed that pyrolysis temperature and oxygen concentration were the most important factors. In contrast, Local Interpretable Model-Agnostic Explanations revealed that oxygen was the most important factor in individual forecast cases. A Monte Carlo simulation with 20,000 samples showed that the projected yield distribution had more than one mode, with pronounced peaks at 20, 35, and 48 wt %. Sobol sensitivity indices showed that hydrogen and pyrolysis temperature were the main factors affecting pyrolysis oil yield, followed by oxygen. Generally, this work offered a complete data-driven plan for predicting the efficiency of pyrolysis systems by combining accuracy, uncertainty quantification, and interpretability.