Abstract
This study presents an integrated framework combining scientometric analysis, machine learning (ML), and explainable artificial intelligence (XAI) to predict the unconfined compressive strength (UCS) of lime and cement-stabilised soils. A century-spanning scientometric review of lime- and cement-based soil stabilisation literature (1912–2026), complemented by a focused scientometric assessment of machine-learning-based UCS prediction studies (2011–2025), revealed a mature yet evolving research domain with strong thematic shifts toward sustainability, binder innovation, and data-driven modelling. A curated dataset of 194 cleaned UCS instances was developed from literature, incorporating soil chemistry, texture, and stabiliser dosage. Four ML models, namely artificial neural network (ANN), random forest (RF), adaboost (AdB), and extreme gradient boosting ( XGB), were developed, with XGB demonstrating the highest predictive accuracy and lowest error metrics. XAI tools, including SHapley Additive exPlanations (SHAP) summary, decision, force, and local interpretable model-agnostic explanations (LIME) plots, and individual conditional expectation (ICE) plots and partial dependence plots (PDP), were applied to interpret model behaviour. Cement and lime emerged as the dominant global predictors, while organic content and pH exerted strong moderating effects; soil gradation variables contributed minimally. The XAI outputs aligned with established stabilisation mechanisms, confirming the physical plausibility of the ML predictions. Overall, this study provides a transparent and mechanistically interpretable ML-based predictive framework that enhances scientific understanding and supports reliable engineering decision-making in soil stabilisation practice.