International Journal of Science, Engineering and TechnologyAuthors: Debashish Chandra, Jasvir Singh
Abstract: Accurate prediction of the unconfined compressive strength (UCS) of cement-stabilized soils is critical for optimizing pavement and subgrade design in geotechnical engineering. Traditional empirical models often fail to capture the complex, nonlinear relationships among influencing parameters such as cement content, curing duration, and soil properties. This study proposes a data-driven framework utilizing two artificial intelligence (AI) models—Random Forest (RF) and Artificial Neural Network (ANN)—to predict UCS based on laboratory and field data collected from diverse Indian soil conditions. Seven input features were considered: cement content, liquid limit, plasticity index, maximum dry density, optimum moisture content, fines content, and curing time. The dataset was preprocessed using Min-Max normalization, and models were trained and tested using a 70:30 split. Performance evaluation using R², RMSE, MAE, IOA, and a20 metrics indicated that ANN slightly outperformed RF, achieving an R² of 0.942 and an a20 of 94.6%. Feature importance analysis revealed that cement content and curing time had the most significant influence on UCS. SHAP analysis further enhanced interpretability of the ANN model. The results demonstrate the reliability and efficiency of AI-based approaches for UCS prediction, offering a robust alternative to conventional methods for soil stabilization design in Indian geotechnical engineering contexts.
DOI: http://doi.org/10.5281/zenodo.17759267
