Geotechnical Performance and Prediction of Fly Ash–Waste Paper Sludge Stabilized Soils via Advanced Machine Learning Techniques

This study develops a predictive framework for estimating the unconfined compressive strength (UCS) of clayey soils stabilized with fly ash (FA) and wastepaper sludge (WPS) using supervised machine learning (ML) models. A total of 80 samples with varying FA (5–25%) and WPS (5–25%) contents were prepared and cured for 3 to 90 days. Geotechnical properties including differential free swell (DFS), liquid limit (LL), plastic limit (PL), optimum moisture content (OMC), maximum dry density (MDD), and UCS were measured as per Indian standard codes. Five regression models—Linear Regression (LR), Support Vector Regression (SVR), Random Forest (RF), AdaBoost, and XGBoost—were trained and tested (80:20 split) in Google Colab using Scikit-learn and XGBoost libraries. RF achieved the highest training accuracy (R² = 0.9989, MAE = 1.72 kPa), while XGBoost closely followed (R² = 0.9979, RMSE = 2.82 kPa). On the test set, RF and XGBoost maintained superior performance (R² = 0.9571 and 0.9423, respectively), with SVR performing the weakest (RMSE = 24.76 kPa). Sensitivity analysis and feature importance identified DFS, curing period, and WPS as key influencers. Surface plots showed optimal UCS (~ 237 kPa) at FA ≈ 15% and WPS ≈ 10–12% after 28-day curing. The study confirms that ensemble models, particularly RF and XGBoost, are highly effective in predicting UCS, offering a reliable and interpretable tool for data-driven geotechnical applications.