Predictive modeling of product yields in microwave-assisted co-pyrolysis of biomass and plastic with enhanced interpretability using explainable AI approaches

Abstract

Microwave-assisted co-pyrolysis of biomass and plastic offers a transformative approach to converting waste into valuable resources such as bio-oil, biochar, and biogas, while simultaneously addressing critical environmental challenges associated with plastic disposal. This research employs explainable AI methodologies to enhance the prediction and analysis of product yields in biomass-plastic co-pyrolysis. Advanced machine learning techniques, including Decision Tree, Random Forest, Extreme Gradient Boosting (XGBoost), and Artificial Neural Networks, were utilized to model yield predictions effectively. The models were fine-tuned through hyper-parameter optimization, achieving high accuracy levels. The study emphasizes the scientific importance of integrating explainable AI with pyrolysis processes to optimize waste-to-resource recovery, contributing significantly to sustainable waste management and circular economy initiatives. Among these, the XGBoost model demonstrated superior performance, achieving R² values of 0.91 for biochar yield, 0.92 for bio-oil yield, and 0.82 for biogas yield on testing sets. To enhance model interpretability, SHapley Additive exPlanations (SHAP) and Partial Dependence Plots (PDPs) were utilized to assess feature importance and examine parameter influences on yield outcomes, offering valuable insights into process optimization and control. Volatile matter and fixed carbon were key predictors for biochar yield, while moisture content and pyrolysis temperature were significant for predicting bio-oil and biogas yields. This study highlights the potential of explainable AI models in advancing sustainable and efficient bio-product recovery from waste materials.