Predicting Traffic Accident Risk at Localized Geographical Areas

This capstone project focuses on the pressing issue of traffic accidents, aiming to enhance our understanding of accident causes and promote safer driving practices. The study’s main objectives are to analyze the relationships between various weather conditions and accident likelihood, assess the influence of traffic management systems on accident risk, and develop a localized machine-learning model for accident probability prediction.

A comprehensive Kaggle dataset from 2016 to 2023 was organized by geographic categories to achieve these goals. Through meticulous preprocessing, distinct subsets were used to train and validate machine learning models employing different classifiers. Five models exceeded established thresholds for sensitivity and area under the curve (AUC). Notably, precipitation and traffic management emerged as crucial factors in accident risk prediction, with gradient boosting and decision tree models showing exceptional predictive capabilities.

These findings suggest the feasibility of accurate accident risk prediction, potentially benefiting navigation systems developed by industry leaders such as Apple, Google, and Waze. The study also highlights the need for future research to expand datasets and variables, providing deeper insights into the advancement of road safety strategies. Overall, this dissertation establishes a strong foundation for future efforts to mitigate traffic accidents and enhance road safety standards.