A comparative analysis of machine learning models for predicting faulting in jointed plain concrete pavements

Abstract

Faulting is defined by variations in elevation at transverse joints in Jointed Plain Concrete Pave- ments resulting from environmental factors, subgrade properties, and traffic loads. It is a major distress for rigid pavements, possessing crucial challenges for maintaining road safety standards. Traditional regression methods often fail to address the complexities of faulting, while machine learning approach utilizes data driven learning to enhance prediction accuracy. Datasets for this study were sourced from the LTPP database, focusing on dry climate zones. Key environmental factors affecting wheel path faulting include Yearly Precipitation, Tempera- ture, Freeze-Thaw Cycles, along with structural properties such as Pavement Thickness, Pavement Age, Tensile Strength, and Optimum Moisture Content are utilized as model input. Five machine learning methodologies, including Support Vector Machine, Decision Tree, Linear Discriminant Analysis, Ensemble and Artificial Neural Network were implemented. Among these, ANN demonstrated highest prediction accuracy, attaining an R² of 0.81. The ANN model was further evaluated to assess the influence of the input variables on the model output through sensitivity analysis.