ABSTRACT
Small-strain shear modulus ([Formula: see text]) of soils is a crucial dynamic parameter that significantly impacts seismic site response analysis and foundation design. [Formula: see text] is susceptible to multiple factors, including soil uniformity coefficient ([Formula: see text]), void ratio (e), mean particle size ([Formula: see text]), and confining stress ([Formula: see text]). This study aims to establish a [Formula: see text] database and suggests three advanced computational models for [Formula: see text] prediction. Nine performance indicators, including four new indices, are employed to calculate and analyze the model's performance. The XGBoost model outperforms the other two models, with all three models achieving [Formula: see text] values exceeding 0.9, RMSE values below 30, MAE values below 25, VAF values surpassing 80%, and ARE values below 50%. Compared to the empirical formula-based traditional prediction models, the model proposed in this study exhibits better performance in IOS, IOA, a20-index, and PI metrics values. The model has higher prediction accuracy and better generalization ability.