Predictive Analysis of Azure Machine Learning for the Rheological Behaviour of Unaged and Polymer Modified Bitumen

Abstract

Rheology can be defined as the primary measurement associated with bitumen flow and deformation characteristics. In the long term, DSR testing consumes a long time, expensive cost and skilled labour to operate equipment or machines in the laboratory. The complex modulus, G* and phase angle, δ, are essential parameters for characterising and predicting the rheological behaviour of unaged bitumen (UB) and polymer-modified bitumen (PMB) in the model. This study developed three regression models using Azure machine learning (AML) to predict the rheological behaviour of UB and PMB. There are three types of data used as input data to develop the regression model: temperature, frequency, and modified material content. Regression models were developed with three processes or steps that need to be prioritised: data collection, model preparation, and model validation. Algorithms used in model development are decision tree regression (DFR), boosted decision tree regression (BDTR) and linear regression (LR). The results show G* and δ values. The R<sup>2</sup> values in the G* and <em>δ</em> predictions obtained from the DFR models are 0.8199 and 0.9480, respectively. Moreover, the R<sup>2</sup> values in the G* and <em>δ</em> predictions obtained from the LR models are 0.4219 and 0.7836, respectively.