Towards Sustainable Pavement Materials: Statistical Modelling of Stability and Flow Characteristics in Recycled PET-Modified Bituminous Concrete

This research explores the use of data-driven modelling approaches to predict the stability and flow behaviour of bituminous concrete mixtures incorporating polyethylene terephthalate (PET) waste as a modifier. A total of 35 unique mix combinations were produced via the dry-mixing technique, varying both PET and bitumen contents, and tested in accordance with MoRTH and ASTM D1559 protocols. Three predictive frameworks, Multiple Linear Regression (MLR), Artificial Neural Networks (ANN), and Random Forest (RF), were developed using a 66:34 train-test data split. Model performance was assessed through the coefficient of determination (R²), mean absolute error (MAE), and root mean square error (RMSE). For stability estimation, RF achieved an R² of 0.9886 on the training set and 0.7902 on the test set, while ANN delivered 0.9916 and 0.9459, respectively. MLR provided a reliable yet lower R² of 0.8683 for the training data. Similar performance patterns were noted for flow value predictions. Although RF demonstrated superior accuracy during training, ANN showed better generalization on test data. MLR remained consistent and interpretable across both datasets. The findings demonstrate the effectiveness of machine learning in streamlining laboratory processes and enhancing the design of PET-modified asphalt mixes, contributing to environmentally sustainable pavement engineering.