Accidental prone area detection in Dhaka metropolitan city using machine learning models

Abstract

Dhaka, known for its status as one of the world's most densely populated areas, grapples with numerous challenges associated with urbanization and safety. This study focuses on identifying accident-prone zones within the Dhaka Metropolitan City by analyzing data collected from 2019 to 2023. Ten specific locations were investigated using various machine learning models: Decision Tree Classifier, Gradient Boosting Classifier, Random Forest, Gaussian Naive Bayes, and Logistic Regression. Evaluation metrics, including precision, recall, accuracy, F1-score, and standard deviation, were utilized, with the Random Forest classifier exhibiting the highest accuracy at 94.39%. The analysis encompassed both monthly and annual accident data, revealing trends and probabilities associated with different incident types. The study's implications suggest potential advancements in accident investigation by incorporating advanced machine learning techniques, integrating diverse datasets, and expanding the geographical coverage to encompass the entire metropolitan area. This comprehensive approach aims to bolster accident prevention measures and enhance urban safety. Overall, this research contributes to a deeper understanding of accident-prone regions and proposes strategies for targeted interventions to improve safety within metropolitan areas