Bayesian estimation of resilient modulus of Jiangsu soft soils from sparse data—Gaussian process regression and cone penetration test data-based modelling and analysis

A Gaussian process regression (GPR)-based model for predicting the resilient modulus of Jiangsu soft soils is developed based on the complied database for Jiangsu soft soils in literatures. The model takes the cone penetration test (CPT) data (e.g., tip resistance q_c data, and sleeve friction f_s data), water content and dry unit weight of soft soils as the input, while provides the predicted resilient modulus as well as quantified uncertainty as the output. By comparing with some conventional machine learning methods, the GPR model can reasonably reflect the correlation between the resilient modulus and the other geotechnical parameters of Jiangsu soft soils. Besides, the GPR model can achieve good performance even when the number of the training dataset is small, which is validated in this study in terms of effectiveness, efficiency and robustness. The GPR method can be considered as a new way for the probabilistic and non-parametric estimation of the resilient modulus of Jiangsu soils.