Quantitative prediction of differential settlement based on machine learning techniques

Differential settlement at the junction of existing and new embankments is a critical challenge in road-widening projects. This study develops a data-driven machine learning (ML) framework to accurately predict this settlement, overcoming the limitations of traditional finite element methods. A comprehensive dataset was generated with finite element simulations (iSight-ABAQUS) and used to train and evaluate eight ML algorithms. Among these, the Gradient Boosting model demonstrated superior performance, achieving an R² value approaching 1.0, indicating exceptional predictive accuracy and generalization. The finalized model was implemented in a Python-based framework, enabling rapid forecasting of road behavior. This ML approach facilitates accelerated design optimization, identifies key influencing parameters like the modulus of elasticity, and supports proactive maintenance planning. The research establishes a transformative, data-driven paradigm for pavement engineering, offering a tool for rapid performance prediction and damage assessment that significantly outperforms conventional simulation-based methods in speed and efficiency.