AI-based algorithms for estimating hydrochar properties in terms of biomass ultimate analysis

• Monte Carlo algorithm is used to find outliers. • 15 different machine learning models are applied for hydrochar yield and HHV. • SHAP analysis pinpointed carbon content as the most influential factor affecting hydrochar HHV. • SHAP analysis pinpointed temperature content as the most influential factor affecting hydrochar yield. Hydrochar, a sustainable biofuel derived from biomass via hydrothermal carbonization, requires precise prediction of its yield and higher heating value (HHV) for efficient utilization. This study uniquely integrates an extensive dataset and a comprehensive suite of state-of-the-art machine learning algorithms and advanced feature importance analysis to enhance the predictive modeling and interpretability of hydrochar yield and HHV based on biomass properties and process parameters. Various machine learning approaches were evaluated for their predictive performance. The Decision Tree model achieved the highest accuracy in yield prediction, with an R² of 0.9445, mean squared error (MSE) of 16.43, and mean relative deviation (MRD) of 2.66 %. The CatBoost model outperformed other algorithms for HHV prediction, yielding an R² of 0.9788, an MSE of 1.63, and an MRD of 3.71 %. All models were rigorously validated using these statistical metrics, supported by visual evaluations such as cross-validation plots and error distributions. Feature importance analysis using SHAP (Shapley Additive Explanations) identified carbon content as the principal determinant of HHV and temperature as the key factor influencing yield. By demonstrating these models' superior accuracy and reliability, this work highlights the feasibility of machine learning-driven predictive frameworks for optimizing biomass conversion into hydrochar.