Towards a Spatially Intelligent Approach for Road Safety Analysis in Tunisia: Spatial Autocorrelation Approach vs Crash Count Models

Road traffic accidents (RTAs) are a significant public health and socioeconomic challenge, particularly in developing countries like Tunisia. This study employs a spatially intelligent approach to road safety analysis by integrating spatial autocorrelation techniques and crash count models to identify and predict accident-prone areas. Using the Local Moran Index, hazardous road locations (HRLs) were mapped along Sousse's major highways, revealing strong spatial clustering of accidents. The Poisson and Negative Binomial (NB) models were applied to assess accident frequency, highlighting key determinants such as driver behaviors (speeding, alcohol/drug influence, failure to respect priority), infrastructure deficiencies, and adverse road conditions. Results indicate that the NB model provides better predictive accuracy due to its ability to account for overdispersion in crash data. The findings underscore the need for targeted safety interventions, including enhanced traffic enforcement, infrastructure improvements, and data-driven policy measures to mitigate accident risks. Future research should explore real-time traffic data integration and machine learning models to refine predictive capabilities further.