Machine learning and design of experiment for enhanced biomass using N2 and CO2-enriched biochar at different pyrolysis temperatures

This study presents the impact of five different biochars prepared at different pyrolysis temperatures, enriched with nitrogen and carbon dioxide as a precursor. The study was developed by utilizing design of experiment (DOE) for response surface methodology (RSM). Different concentrations of biochar were added to the liquid culture medium for in vitro regeneration of Ceratophyllum demersum, which revealed the better performance of biochar prepared at low temperatures on plant biomass and pigmentation. Results analyzed by correlation analysis exhibited a significantly negative impact of biochar concentration and type on plant biomass and biochemical parameters, underscoring the intricate relationship within plant antioxidative mechanisms and growth response. Different machine learning (ML) models namely Random Forest (RF), Multi-Layer Perceptron (MLP), Gaussian Process Regression (GPR), Extreme Gradient Boosting (EXGB), CatBoost Regressor (CBR), and LightGBM Regressor (LGBMR) were employed for predictive accuracy and robustness using the coefficient of determination (R2) and Mean Squared Error (MSE). Results indicated that RF, XGB, and CBR consistently achieved superior predictive performance for chlorophyll-a (Chl-a), chlorophyll-b (Chl-b), and total chlorophyll (Chl-T), tannic acid (TA), with high R2 and low MSE values. The RF model used for the Y-Permutation test, feature importance test, and SHAP test confirmed the predictive performance. Feature importance test figured out concentration, while SHAP analysis illustrated biochar type as the most significant factor despite close scores. These insights may help to formulate next-generation biochar production for sustainable crop production and wastewater-based plant systems under climate stress conditions.