From theory to deep learning: Understanding the impact of geographic context factors on traffic violations

Abstract

Traffic violations pose significant threats to road safety, often leading to accidents, injuries, fatalities, and property damage. While driving behavior results from a dynamic interaction among drivers, vehicles, and the environment, most studies have primarily focused on human factors, neglecting the critical influence of social and environmental elements of the geographical context. To address this gap, we first develop a theoretical framework that explores how vehicles, drivers, and environmental factors interact, with particular emphasis on the environmental impact on driving behavior and road safety. Building on this framework, we introduce TraVio, a multitask deep learning model that combines Graph Convolutional Networks (GCNs) with Multilayer Perceptrons (MLPs) to detect and classify traffic violations. Through a case study in Montgomery County, Maryland, we demonstrate that TraVio outperforms traditional machine learning baselines in both binary and multi-label classification tasks. The results underscore the critical role of geographic context and time factors in shaping traffic violations. This study not only provides a robust modeling technique for predicting traffic violations but also offers a theoretical foundation that can be extended to other driving-related themes, contributing to safer road environments through informed urban planning and policymaking.