Experimental investigation of coal bottom ash concrete mechanical properties and development of novel swarm optimized tree-based explainable models

The thermal power plants generate a significant amount of coal bottom ash (CBA), which poses environmental hazards. Effective utilization of CBA is essential to mitigate its adverse environmental impacts. In this study, the feasibility of using CBA-from two coal-fired power plants in Punjab, Pakistan, namely Sahiwal (SWL) and Sheikhupura (SKP)-as a partial sand replacement in concrete was investigated to promote sustainable concrete production. A series of concrete mixtures and experiments were conducted, including evaluations of Flexural Strength (FS) and Split Tensile Strength (SPT) at curing ages of 7, 28, and 90 days for M25 and M35 grade concrete. To enhance predictive capabilities, eXtreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LGBM) models optimized using Particle Swarm Optimization (PSO) were employed to accurately predict the FS and SPT of CBA-based concrete. In addition to experimental data, previously published datasets were incorporated to improve model robustness. Experimental results indicated that concrete incorporating SWL-CBA achieved FS and SPT values comparable to control mixes at 28 and 90 days for M25 and M35 grades when CBA replaced 25 % and 50 % of sand, respectively. However, SKP-CBA mixes consistently showed lower strength performance. The PSO-optimized XGBoost and LGBM models exhibited excellent predictive performance, with R² values exceeding 0.98 during training and reaching 0.96 during testing. Furthermore, SHapley Additive exPlanations (SHAP) analysis was used to interpret the PSO-LGBM model, revealing that curing age, the specific gravity (SG) of CBA, sand/cement ratio, and CBA, and curing age, cement content, water and CBA were the most influential features affecting FS and SPT, respectively. The study findings support the recommendation of using higher SG CBA as a sustainable partial sand replacement, contributing to natural sand conservation, while also highlighting the effectiveness of machine learning approaches in accurately modeling and optimizing concrete performance.