Machine learning-based prediction of resilient modulus for blends of tire-derived aggregates and demolition wastes

Abstract

Although numerous studies have been undertaken to examine the potential of C&D wastes in pavement structures, limited research to date has developed constitutive models for the resilient modulus (Mr) of C&D wastes. In this study, machine learning algorithms: artificial neural network (ANN); Support Vector Regression (SVR); and Random Forest Regression (RFR) were utilised for estimating the Mr of C&D wastes blended with tire-derived aggregate (TDA). A database of repeated load triaxial test results on C&D/TDA blends including stress-state parameters, Mr values and other influential parameters on the Mr such as physical properties of the blends, as well as the California bearing ratio (CBR) test results was utilised for developing the machine learning models. The predictive performance of the developed algorithms was assessed using statistical indices and several validation stages. The results exhibited the robustness of machine learning algorithms used for predicting the Mr of C&D/TDA wastes, with R 2 train, R 2 test, MAPEtrain, MAPEtest between 0.996–0.997, 0.911–0.997, 0.037–0.068, and 0.037–0.067, respectively. The sensitivity analysis results indicated that the optimum moisture content and CBR had the highest impacts on the Mr of C&D/TDA blends.