Machine learning based studies on settlement characteristics of soft clayey soil

Understanding and predicting the settlement characteristics of clayey soils is critical for ensuring the stability and performance of geotechnical structures. Traditional methods for settlement estimation often rely on empirical correlations and time-consuming laboratory tests, which may lack precision when applied to complex soil behaviour. This study utilizes machine learning (ML) techniques, specifically Support Vector Machines (SVM) and Random Forest (RF) models, to predict settlement characteristics of clayey soils. Key input parameters include soil consistency limits, unit weight, initial void ratio, pre-consolidation pressure, and loading conditions. The dataset was compiled from experimental tests and validated against established settlement models. The performance of the SVM and RF models was evaluated through metrics such as mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²). The findings demonstrate that the Support Vector Machine (SVM) model outperformed the Random Forest (RF) model in predicting settlement behaviour, achieving a high R² value of 0.9893, minimal error metrics (MAE: 0.0034, RMSE: 0.0040), and a low MAPE of 1.1816%, indicating superior predictive accuracy.