A Comparative Analysis of Crowdsourced and Kernel Density Approaches for Improved Accidents Blackspots Prediction Accuracy

The accurate identification of accident blackspots is indeed critical for implementing effective road safety measures. Blackspots, being areas with a higher incidence of accidents, demanding focused attention to mitigate risks and enhance overall road safety. This study investigates methods to improve the accuracy of predicting accident blackspot locations in a case study on the roads of Sistan Baluchistan province in Iran. A comprehensive dataset spanning five years of meticulously recorded accident records was collected in collaboration with on-duty traffic police officers. The research employs binary logit models to identify significant variables contributing to blackspot prediction accuracy. Noteworthy factors encompass driver attributes (such as age, gender), road features, environmental elements, weekday/weekend status, road type and traffic volume. Comparative analyses of individual implementations of Kernel density and Crowdsourcing methods revealed prediction accuracies of 62.3% and 65.3%, respectively. However, when these methods were jointly applied to extract common blackspots, the prediction accuracy significantly increased to 70.02%. This combined approach showcased the synergistic potential of utilizing diverse methodologies, emphasizing the necessity of integrating multiple data sources for precise accident blackspot identification. The findings underscore the effectiveness of amalgamating Kernel density and Crowdsourcing techniques, offering a promising avenue for enhancing predictive models and implementing proactive road safety measures within the diverse road networks. The combined method pinpoints high-risk locations validated by both crash data and driver behavior, giving road authorities a reliable tool to prioritize safety interventions with higher confidence.