Surrogate measurement of traffic safety with insights from pavement condition, sensor fusion, and machine learning.

Traffic safety is one of the crucial components of transportation management. Annually, many individuals lose their lives in traffic crashes. Therefore, identifying factors influencing traffic safety holds significant importance. The four main factors affecting traffic safety are human, vehicle, road, and environmental conditions. As a subset of road conditions, pavement conditions directly and indirectly impact traffic safety by influencing drivers' behavior and vehicle movement. Thus, evaluating pavement conditions and their effects on drivers can significantly improve road conditions and enhance traffic safety. Despite the impact of pavement conditions on traffic safety, this aspect has received less attention from researchers. Accordingly, the primary objective of this study is to investigate the influence of pavement deterioration on aggressive driver behavior and traffic safety. In this research, utilizing a developed system, data were collected using smartphone sensors to predict aggressive driver behaviors and cameras to capture road surface conditions. These data were evaluated using artificial neural networks and computer vision models. These models accurately predicted drivers' behavior with an F1 score of 91.6%, ride quality defects with an F1 score of 97.2%, and pavement defects with an mAP of 70.7%. Finally, the collected data and developed models formed a dataset of pavement defects and aggressive driver behaviors resulting from these defects. Random Forest and Support Vector Machine (SVM) were utilized to examine the impact of pavement deterioration and ride quality defects on aggressive behaviors. These models successfully predicted drivers' behavior with an accuracy of 96.6% and 96.7% using pavement defects. This study contributes a novel framework integrating machine learning with statistical validation to uncover specific correlations between pavement defects and aggressive driver behaviors. This research provides valuable insights for targeted road maintenance and traffic management strategies by pinpointing the most influential types of pavement deterioration on aggressive driving.