Multi-task deep learning for joint prediction of traffic emissions and travel delay

Abstract

Signalised intersections play a crucial role in urban traffic management, ensuring the smooth movement of vehicles across road networks. However, urban intersections are often hotspots for congestion, increasing emissions, extending travel delay, and posing challenges for sustainable operations of traffic. The existing traffic management methods typically focus on either travel delay or emissions in isolation, neglecting their inherent interdependence; congestion simultaneously increases emissions and travel delay. This study introduces a novel deep learning framework termed multi-task temporal convolutional network (MT2CN) that jointly predicts traffic emissions and travel delay at signalised intersections. It is evident from our findings that the proposed MT2CN approach outperforms the conventional single-task models, indicating a significant finding for predictive modelling. By utilising advanced deep learning techniques and explainable artificial intelligence techniques, such as Shapley additive explanations (SHAP), our framework provides more accurate predictions and explainable insights to facilitate sustainable and intelligent traffic management.