Optimization of pond-ash-based controlled low-strength materials with lime and superplasticizer via experiments and supervised machine learning

• Evaluated the strength and flowability characteristics of high-strength CLSM using various binding agents cement, lime, and superplasticizers along with pond ash and water. • Machine learning models (XGBoost, XGBoost-GWO, XGBoost-PSO, and XGBoost-SSO) predicted UCS of CLSM accurately. • XGBoost-SSO model showed the highest accuracy with R² = 0.990 for training, 0.974 for testing. • Pond Ash content, cement, and SP content most influenced UCS, per sensitivity analysis. • Optimum mix: 9 % Cement, 2 % lime; met criteria for high-strength CLSM. The growing production of industrial byproducts such as pond ash and fly ash from thermal power plants presents a major waste management challenge. Integrating these byproducts into controlled low-strength materials (CLSM) offers a sustainable solution for backfilling behind retaining walls, tunnels, and utility trenches. For applications requiring higher strength, CLSM mixtures must achieve compressive strengths above 0.7 MPa after 28 days. This study investigates the effects of adding superplasticizers and lime to conventional CLSM materials through experimental work to develop high-strength CLSM mixtures. Results show a significant improvement in compressive strength with these additives, a finding not previously reported. To complement the experiments, machine learning models were developed to predict the unconfined compressive strength (UCS) of CLSM based on varying proportions of cement, lime, superplasticizers (SP), and pond ash as traditional experimental and empirical approaches are limited in capturing nonlinear interactions among mix parameters. A comprehensive dataset was created from systematic variations in mix proportions and corresponding strength measurements. Four predictive models XGBoost, XGBoost-GWO, XGBoost-PSO, and XGBoost-SSO were trained and tested. The XGBoost-SSO model achieved the best performance with R² values of 0.990 (training), 0.979 (validation), and 0.974 (testing), along with the lowest RMSE (0.026 MPa) and MAE (0.019 MPa) in the testing phase. Regression and REC analyses confirmed its superior predictive capability. Sensitivity analysis identified pond ash (55 %) and cement (17.6 %) as the most influential factors. A user-friendly GUI tool was also developed for real-time UCS prediction and data-driven mix optimization.