Bayesian optimization-guided machine learning models for bio-oil yield prediction from lignocellulosic biomass pyrolysis

The transition to renewable energy needs efficient biomass conversion technologies. Pyrolysis converts lignocellulosic biomass into bio-oil, but optimizing yield remains challenging due to complex nonlinear relationships. The present study introduces a novel hybrid framework integrating Bayesian optimization with five machine learning algorithms (CatBoost, Random Forest, K-Nearest Neighbors, Artificial Neural Networks, and ElasticNet) to enhance bio-oil yield prediction. Relying on a dataset of 329 data points encompassing biomass composition and pyrolysis conditions, Bayesian optimization efficiently tuned hyperparameters, overcoming limitations of traditional methods. The CatBoost model outperformed others, achieving a Mean Absolute Error (MAE) of 2.251±0.354 and R²=0.861±0.061. Feature importance and partial dependence analysis identified Hemicellulose (10–15 wt%) and Heating Rate (700–800 °C/min) as critical variables for yield maximization. The developed work demonstrates the transformative potential of Bayesian-optimized ML for sustainable bioenergy production.