Comparing Machine Learning Models with Witczak NCHRP 1-40D Model for Hot-Mix Asphalt Dynamic Modulus Prediction

The design and construction of a structurally and functionally stable pavement are pivotal for sustainable mobility. The need for a structurally stable and flexible pavement involves the assessment of the various engineering properties of asphalt. The use of the Witczak model is useful in assisting pavement designers with limited laboratory tests in the estimation of asphalt concrete dynamic modulus (E*). This is because the existing regression and artificial neural networks (ANN) model training using Witczak model input parameter has not exceeded 91% correlation between the measured and predicted E* and the huge error which could constitute a significant increase in pavement cost. In this research, five machine learning models were used to model E* and Log E*. To achieve the aim of this research, Witczak Model was adopted. Witczak model was used to input the obtained parameters and the database containing 7400 data points. The performance of the machine learning models was compared with the Witczak model. A global sensitivity analysis (GSA) was carried out to ascertain the model parameter importance to the output variance using the easyGSA MATLAB tool. The results of the research revealed that the Gaussian process regression (GPR) have a high predictive capability, with the highest coefficient of determination (R-2) of 0.95 and 0.93 for E* and Log E*, respectively. The results strongly suggest that the GPR model could be used as an alternative to Witczak regression and ANN models. The GSA results showed that the gradation, volumetric properties and the phase angle have a significant effect on the E* prediction where the volumetric properties and cumulative weight retained on the 1.9 cm sieve induced the maximum effect on the prediction of Log E*. The outcome of this research will be of immense benefit to transportation engineers, highway engineers, researchers and construction workers on the use of this model for the prediction of the dynamic modulus of flexible pavement for sustainable mobility.