Machine learning models to predict mechanical performance properties of modified bituminous mixes: a comprehensive review

Abstract

The incorporation of various modifiers such as rubber, plastic, fibers, and anti-stripping agents has demonstrated favourable effects on the mechanical properties of bituminous mixes, including Marshall stability (MS) and indirect tensile strength (ITS), thereby addressing various challenges associated with conventional bitumen. Recent research has notably focused on predicting the mechanical performance of both unmodified and modified bituminous mixes using advanced machine learning (ML) techniques, offering potential solutions to issues encountered in classical laboratory experiments. The present comprehensive review synthesizes the existing literature on ML techniques for predicting MS and ITS of bituminous mixes. Initially, it reviews the range of inputs utilized and suggests missing inputs. The impact of optimal user-defined parameters on discrete model performance, along with model comparison relying on statistical metrics, is analysed to recognize ML models with adequate predictive potential. Additionally, the paper examines the validation aspect of the model dataset in terms of experiments, providing insights for model developments in future. Overall, this study aims to deliver an overview of the present status of ML models for predicting MS and ITS, highlighting research gaps for model development and attaining anticipated performance. Hence, the condensed knowledge will prove invaluable in directing future research efforts towards the development of sustainable and efficient modified bituminous mixes.