With the accelerated deployment of connected and automated vehicle (CAV) technologies, public agencies have urgent needs on how to utilize these rich data sources of CAVs to improve traffic mobility, safety, and environmental and energy impact. This research will tackle one of the big data challenges, which is mining driving behavior patterns using vehicle data sources. We leverage physics-informed deep learning and uncertainty quantification methods to predict drivers’ car-following behavior using historical trajectories. A digital twin is developed leveraging the COSMOS testbed deployed near Columbia campus to validate the model algorithms and results. Moreover, an app is developed that captures drivers' faces. On the AWS server, face detection algorithms are applied to analyze drivers' moods and attention. Combined with the vehicle information (e.g., speed, acceleration) that is detected from roadside cameras, a model is established to predict the safety index of the driver and the roadway. The project outcome will be valuable for digital sibling simulation development and applications and future deployment of AVs that need to drive alongside humans.