Uncertainty-Aware Prediction of Bearing Capacity of Shallow Foundations Resting on Cohesionless Soils Using Bayesian Regression

This paper addresses the critical gap of uncertainty quantification in existing geotechnical models predicting bearing capacity for shallow foundations on cohesionless soils. The developed Bayesian regression model improves point estimate accuracy by 30% compared to traditional models while introducing a robust method for quantifying prediction uncertainty. Utilizing extensively studied literature data, we demonstrate the model’s closed-form structure, allowing the direct calculation of both the most probable bearing capacity and its uncertainty. This transparency distinguishes the proposed model from black-box machine learning alternatives. Furthermore, we illustrate its practical applicability by integrating it into reliability-based design, specifically showcasing its utility in incorporating applied load distributions. In essence, this research enhances predictive accuracy and establishes a comprehensive framework for addressing uncertainty in bearing capacity models, with direct implications for reliability-based design practices.