Predictive Modeling of Urban Traffic Accident Severity in Türkiye's Centennial: Machine Learning Approaches for Sustainable Cities

Abstract

With their far-reaching implications for public health, urban development, and societal harmony, traffic accidents remain a global challenge. As the Republic of Türkiye marks its 100th year, predicting traffic accident severity assumes critical significance, aligning with the nation's aspirations for urban renewal and sustainable progress. This research harnesses the capabilities of machine learning (ML) to anticipate accident severities, shedding light on the critical roles of specific driver and vehicle characteristics. In-depth evaluation of various ML techniques—spanning from Random Forest (RF) and Gaussian Naive Bayes to k-NN, CatBoostClassifier, LightGBM, and Decision Trees—was undertaken, drawing on an expansive dataset that mirrors a spectrum of traffic situations. The RF algorithm demonstrated superior predictive prowess, with certain variables such as Engine_Capacity_(CC), Age_of_Driver, Age_of_Vehicle, Day_of_Week, and Vehicle_Type emerging as decisive factors in accident outcomes. Beyond highlighting RF's potential in accident severity prediction, the study emphasizes the significance of critical determinants. These insights offer a roadmap for stakeholders to craft specialized interventions, amplify public awareness efforts, and pioneer infrastructural upgrades, culminating in a vision of enhanced road safety. Furthermore, this investigation charts a course for Türkiye to foster a sustainable urban trajectory through informed urban and traffic planning initiatives.