Improving the interpretation of undrained shear strength from piezocone penetration tests by integrating soil physical properties using a hybrid meta-heuristic algorithm

Conventional empirical equations for estimating undrained shear strength ( s u ) from piezocone penetration test (CPTu) data, without incorporating soil physical properties, often lack the accuracy and robustness required for geotechnical site investigations. This study introduces a hybrid virus colony search (VCS) algorithm that integrates the standard VCS algorithm with a mutation-based search mechanism to develop high-performance XGBoost learning models to address this limitation. A dataset of 372 seismic CPTu and corresponding soil physical properties data from 26 geotechnical projects in Jiangsu Province, China, was collected for model development. Comparative evaluations demonstrate that the proposed hybrid VCS-XGBoost model exhibits superior performance compared to standard meta-heuristic algorithm-based XGBoost models. The results highlight that the consideration of soil physical properties significantly improves the predictive accuracy of s u , emphasizing the importance of considering additional soil information beyond CPTu data for accurate s u estimation.