Machine learning approaches for estimating the strength performance of recycled aggregate concrete in rigid pavement systems.

Abstract

This study aims to develop a method for predicting the 28-day compressive strength of recycled aggregate concrete (RAC) for rigid pavement applications. A hybrid dataset of 385 observations, combining laboratory results and selected literature data, was used to develop and compare machine learning models. The models were assessed using five-fold cross-validation, error measures, bias analysis, multicollinearity assessment, and SHAP interpretation. Lasso Regression provided the best performance, with R² = 0.7713, MAE = 3.51 MPa, and RMSE = 4.47 MPa. The study is limited to five-fold cross-validation without independent external validation. Its originality lies in evaluating a pavement-oriented hybrid RAC dataset using predictive and interpretive analyses. The results show that regularized models can support preliminary RAC mix evaluation and material assessment.