Experimental Investigation and Machine Learning-Based Prediction and Optimization of Mechanical Properties of Biochar-Enhanced High-Strength Concrete

Biochar has emerged as a sustainable additive in concrete production, offering potential for improved concrete performance and waste valorization. An experimental investigation was conducted using wood waste biochar as a partial cement replacement at 0%, 2%, 4%, and 6% by weight. Compressive strength (CS) and split tensile strength (STS) were determined at 7 and 28 days, while flexural strength (FS) was determined at 28 days. The experimental results demonstrated that 2 to 4% biochar replacement enhanced CS by 9.67% and FS by 15.40%, while STS showed optimal improvement at 2% replacement by 6.24%. To extend these findings across diverse feedstocks and mix designs, a comprehensive database of 318 mixes incorporating 13 biochar types was compiled from literature to develop machine learning (ML) models for predicting all three strength properties simultaneously. Random Forest (RF) and Gradient Boosting (GBR) algorithms were optimized using nested 5-fold cross-validation and compared against a Ridge regression baseline. The optimized RF model ( n _estimators = 1000) achieved a nested cross-validated R 2 of 0.817 ± 0.072 and a 32.5% reduction in RMSE compared to the baseline, with testing R 2 values of 0.894 for CS, 0.828 for FS, and 0.537 for STS. (SHapley Additive exPlanations) (SHAP) analysis identified cement content, coarse aggregate (CA) content, and biochar dosage as the most influential features. Biochar effect curves, based on the most reliable datasets (rice husk, n = 69; wood, n = 52), demonstrated that rice husk biochar consistently enhanced all three strength properties, while wood biochar showed superior performance for FS and STS. Experimental validation using wood waste biochar confirmed that model predictions closely matched measured strengths, with 90% prediction intervals reliably encompassing experimental values. The developed models offer a practical decision-support tool for sustainable concrete mix design, significantly reducing experimental effort while providing evidence-based guidance for biochar feedstock selection and dosage optimization, keeping the cement usage at a minimum.