Machine learning framework for predicting and improving the unconfined compressive strength and california bearing ratio of lateritic soil stabilized with industrial wastes

Abstract Industrial waste materials are increasingly used in geotechnical engineering as partial replacements for cement, offering cost-effective and environmentally sustainable alternatives. This study investigates the California Bearing Ratio (CBR) and unconfined compressive strength (UCS) of lateritic soil stabilized with red mud (RM), copper slag (CS), and iron ore tailings (IOT) in proportions of 5–45%. A systematic laboratory program generated 155 experimental datasets, which were further used to develop predictive models with machine learning algorithms including K-Nearest Neighbours (KNN), Decision Tree Regressor (DTR), Random Forest Regressor (RFR), and Multi-Layer Perceptron (MLP). Statistical indices the coefficient of determination (R 2 ), root mean square error (RMSE), and mean absolute error (MAE) along with Taylor diagrams and Regression Error Characteristic (REC) curves were applied for model evaluation. RFR and MLP achieved R 2 values above 0.90, showing superior performance. SHAP (SHapley Additive exPlanations) analysis highlighted curing period, maximum dry density (MDD), and CS dosage as the most influential features. Results confirmed that 30% CS significantly enhances both UCS and CBR, demonstrating its potential as a supplementary stabilizer. The study contributes a robust experimental machine learning framework that not only predicts UCS and CBR with high accuracy but also provides mechanistic insights, supporting circular economy practices and low-carbon pavement design.