Jordan continues to face a serious public health challenge due to pedestrian-related traffic accidents, which contribute significantly to human and economic losses. This study aims to investigate the historical characteristics of pedestrian accidents in Jordan and to develop predictive models for pedestrians’ accidents using both classical regression techniques and artificial neural networks (ANNs). Pedestrian accident data were collected from official annual reports published by the Jordan Traffic Institute (JTI) for the years 1984 through 2022.

The analysis revealed that although the fatality risk has decreased over the years from 13 to 6 deaths per 100,000 inhabitances, white it remains relatively high compared to global standards. Children aged 3–5 and elderly individuals over 60 were identified as the most vulnerable age groups. Pedestrian accidents were most frequent during peak evening hours and at low-speed limits of 40–60 km/h, often due to driver negligence or inadequate pedestrian infrastructure.

Three classical regression models to predict pedestrians’ accidents as function of registered vehicles were developed: linear, logarithmic, and power with R² values of 0.937, 0.913, and 0.928, respectively. The linear model showed the best fit among the traditional approaches. Additionally, an ANN model with two hidden layers was trained using registered vehicle data as input, achieving an R² value of 0.981, indicating superior predictive performance and the ability to capture complex nonlinear trends.

These findings highlight the critical role of advanced machine learning techniques in enhancing traffic safety planning and policy formulation. The study recommends integrating AI-driven models into national traffic monitoring systems and adopting urban planning strategies that prioritize pedestrian safety to reduce accidents consequences.