A new data-driven model for vehicle and pedestrian safety: statistical approach based on spatial decision-making

Minimizing the losses that occur after traffic accidents is a primary duty for all humanity. To do so, it is necessary to examine and analyse the potential risk factors that affect the severity of traffic accidents. In this article, a new spatial decision-making-based statistical solution methodology is proposed to determine the accident risk factors that occur in three different accident types using 5-year (2015-2019) accident data. (i) 22 independent variables and 157 sub-variables were determined for the traffic accident categories where vehicle-vehicle, vehicle-pedestrian and vehicle-other collision types occurred, (ii) the fuzzy simple weight calculation method was preferred to determine the effects of risk factors on accident categories, (iii) spatial analyses of risk factors were provided via geographical information system and combined with the obtained effect values, (iv) the current effect of risk factors on accident categories was tested with the multinomial logistic regression model. The multinomial logistic regression model results revealed a strong model fit (McFadden R2 = 0.749) and identified the variables that significantly increase or decrease the probability of each crash type compared to the reference category. For instance, while the geo-intersection had the highest effect for vehicle-vehicle crashes, the pedestrian defect had the highest impact for vehicle-pedestrian crashes. Spatial analysis results also showed that accident severity tends to be higher in the western, southern, and central regions of Türkiye. The proposed methodology offers a comprehensive framework that supports evidence-based policy development for improving traffic safety. The resulting findings serve as a guide for local administrators, policy makers, and traffic safety experts with regard to vehicle and pedestrian safety.