Probabilistic Analysis of Pile Foundation in Cohesive Soil

Pile foundation is an essential part of a structure which fulfills the load bearing requirements in the field of geotechnical engineering. Previous studies that used deterministic approaches in calculating the allowable load carrying capacity have revealed significant uncertainties. To address these uncertainties and improve accuracy, this study utilized machine learning (ML) models to suggest high-performance models for determining the allowable load carrying capacity of pile foundation. Three ML models, namely decision tree (DT), random forest, and adaptive boosting (AdaBoost), were proposed and evaluated using two datasets (70% training and 30% testing) with input variable parameters including diameter (d), cohesion (c), adhesion factor (α), and factor of safety. To authenticate the accuracy of the ML models, different statistical performance parameters were evaluated for both training and testing datasets. These parameters include trend measuring parameters (R², VAF, LMI, a-20 index and KGE) and error measuring parameters (RMSE, NMBE, MAE, MBE, and MAD). Results showed that the DT model had the best outcomes among the three ML models. Additionally, the First-Order Second Moment was used to compute the reliability index (β) and probability of failure (P_f). Regression curves were plotted to gauge the prediction capability of different ML models; the uncertainty analysis was also performed to estimate the reliability of the ML models, while Williams plots were generated to identify the suitable region of the model. Sensitivity analysis is performed to check the impact of different input parameters on the output parameter. Overall, the study demonstrated that ML models can be used to improve the accuracy of determining the allowable load carrying capacity of pile foundation.