Machine learning-based evaluation of indirect tensile stiffness modulus of fiber-modified cold asphalt mixtures

This study aims to evaluate the Indirect Tensile Stiffness Modulus (ITSM) of fiber-modified cold asphalt mixtures using advanced machine learning models. Several asphalt mix designs (123 mixes) from many previous studies incorporating various fiber types, contents, and curing conditions were experimentally tested. Artificial Neural Networks (ANNs) and Deep Neural Networks (DNNs) were developed to predict ITSM based on seven input features, including fibre characteristics and mix design parameters. The ANN model (ANN-I) demonstrated superior performance, with a correlation coefficient (R) of 0.951 and an RMSE of 174.85 MPa, outperforming the DNN model (DNN-II), which showed lower predictive accuracy (R = 0.884, RMSE = 252.56 MPa). Curing time emerged as the most influential variable across both models. These conclusions verify that ANN provides a stronger and more generalizable means of modelling the stiffness behaviour of cold asphalt reinforced with fibres and provide evidence of its effectiveness when optimising sustainable pavement designs.