Sustainable prediction of compression index in fine-grained soils using machine learning techniques

Abstract The accurate prediction of ultimate settlement is of critical importance for the longevity of structures, as it helps prevent structural damage and functional deterioration. Settlement, a function of the soil’s compression index ( C c ), is typically determined through oedometer tests. This research seeks to enhance the prediction accuracy of C c within a sustainability framework by employing advanced machine learning techniques. A series of soil samples with varying clay contents were synthesized by incorporating bentonite into silt, resulting in nine distinct soil classes ranging from low plasticity silt (ML) to high plasticity clay (CH). Consolidation tests were performed on reconstituted samples subjected to 100 kPa pre-consolidation pressure, revealing significant correlations between the compression index and various physical properties. Leveraging these correlations, a comprehensive dataset was compiled, integrating current experimental data with relevant studies from the literature. The predictive capabilities of several machine learning models, including Support Vector Regression (SVR), Artificial Neural Networks (ANN), Random Forest (RF), and Multiple Regression Analysis (MRA), were evaluated. Among these, the RF model exhibited superior performance, achieving an R 2 of 0.83. The study introduces a novel prediction framework that extends beyond conventional settlement estimation by incorporating a comparative ML-based modeling approach supported by both experimental and literature-based data. Furthermore, the methodology demonstrates its practical value by offering a faster and less resource-dependent alternative to traditional consolidation testing, contributing to sustainability oriented and cost-efficient geotechnical design.