Combining-Deeb-Learning-with-Uncertinity

This repository provides the implementation of a hybrid machine learning framework that combines deep neural networks (DNN) and Gaussian process regression (GPR) to predict the rheological properties of asphalt binders and mastics. The model targets the complex shear modulus and phase angle using material composition, aging conditions, and rheological test parameters as inputs. DNN are employed to capture highly nonlinear relationships, while GPR refines predictions and enables uncertainty quantification, improving reliability when data are limited or noisy. The framework is applied in a multiscale manner, linking binder- and mastic-level behavior, with synthetic mastic data generated using finite element modeling. The proposed hybrid approach demonstrates high predictive accuracy and provides a robust, interpretable tool for asphalt material characterization and pavement engineering applications.