Porosity/Cement Index and Machine Learning Models for Predicting Tensile and Compressive Strength of Cemented Silt in Varying Compaction Conditions

This study investigates the mechanical response of cemented silt subjected to 28 days of curing by integrating two predictive methodologies: porosity–cement index (η/Civ) and machine learning (ML) models. The soil was compacted over a wide range of molding water contents and dry densities, including optimum and off-optimum states, and stabilized with varying cement contents. Unconfined compressive strength (qu) and splitting tensile strength (qt) were evaluated as functions of cement dosage, curing time, porosity, water content, and the specific gravities of the soil and cement. The η/Civ index demonstrated a strong predictive capability for both qu and qt, with determination coefficients exceeding 0.980, and exhibited the expected power-law decay with increasing η/Civ. ML algorithms—particularly Gaussian Process Regression with a Matern 5/2 kernel—outperformed the empirical model, achieving R2 values of 0.963 (validation) and 0.997 (testing) for qu prediction. The qt model similarly reached R2 = 0.984–0.988, demonstrating high generalization and stability across curing and compaction conditions. Experimental results revealed substantial strength gains with decreasing η/Civ, with qu increasing from 100 kPa at η/Civ = 46 to 2900 kPa at η/Civ = 19, while qt rose from 10–15 kPa to 300 kPa across the same range.