Advanced modeling techniques using hierarchical gaussian process regression in civil engineering

Abstract

Gaussian process regression (GPR) models, with their desirable mathematical properties and outstanding practical performance, are increasingly favored in statistics, engineering, and other domains. Despite their advantages, challenges arise when applying GPR to extensive datasets with repeated observations. This study aims to develop models for predicting Finland's soft-sensitive clays’ undrained shear strength (Su). The study presents the first correlation equations for Su of Finnish clays, derived from a multivariate dataset compiled using field and laboratory measurements from 24 locations across Finland. The dataset includes key parameters such as Su from field vane tests, reconsolidation stress, vertical effective stress, liquid limit, plastic limit, natural water content, and sensitivity. The GPR model demonstrated high accuracy, with a mean squared error (MSE) of 0.11% and a correlation coefficient (R²) of 0.98, indicating excellent predictive performance. These findings highlight the strong interactions between Su, consolidation stresses, and index parameters, establishing a robust foundation for practical GPR implementation. The GPR model is recommended for forecasting Su due to its high learning performance and ability to display prediction outputs and intervals. This research has significant implications for various civil engineering applications, including transportation, geotechnical, construction, and structural engineering, offering a valuable tool for improving engineering practices and decision-making.