Performance of traditional and machine learning-based transformation models for undrained shear strength

Abstract In geotechnical engineering, transformation models are often used as first estimates of parameters and to verify the order magnitude field laboratory tests, which reliability might be constrained by many uncertainties. The undrained shear strength has been for long particular interest such models. traditional rather simple. Still, community does not seem have agreed upon use. particular, question including index properties seems open. paper, performance is compared that machine learning (ML)-based addition, influence data coherence studied using two datasets different quality. ML-based proved perform better than ones both datasets. Clearly, most dominant variables in model preconsolidation pressure effective vertical stress. Although additional variable may well improve training set, prediction testing sets generally tends worsen, indicating overtraining. risks overtraining increase with incoherent data.