Sustainable Valorization of Biomedical Waste: Predictive Modeling of Thermochemical Product Distribution Using Hybrid Machine-Learning Models

This research concentrates on modeling product distributions in the thermochemical process of biomedical waste processing using a hybrid machine learning model (ensemble of XGBoost and LightGBM). This study focuses on accurately predicting and optimizing the production of hydrothermal, gasification, and pyrolysis products (e.g., bio-oil, syngas, and biochar) using process parameters such as temperature, heating rate, pressure, and pH. Modeling of process parameter–product yield interactions via a hybrid ML framework is carried out. A training dataset was prepared using experimental values along with published data to analyze the parameters in detail. Training datasets were used to form a Pearson correlation matrix that has positive and negative coefficients. A DSS was developed to support the optimization of operating parameters to improve the product yield. A dataset of 180 points was compiled (150 from literature + 30 independent laboratory hydrothermal experiments). The proposed hybrid model achieved a classification accuracy of 94 percent (macro-F1: 0.94) for process selection and regression performance of MAE = 4.6 percent [3.3–6.0], RMSE = 5.9 percent [4.3–7.5], and R2 = 0.78 [0.62–0.88] on the external test set. The study also shows that the feature importance and sensitivity analyses ranked the temperature, catalyst concentration, and residence time as the dominant factors that are influencing the yields. Key factors received priority rankings to generate predictions that were evaluated on other dataset samples. Hybrid ML models can be more suited for highly parameter-oriented processes such as pyrolysis of biomedical waste.