RISK PREDICTION OF TRAFFIC ACCIDENT USING MACHINE LEARNING
Loading...
Date
2022-07-21
Authors
Amani Mohammed Hakawati
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Introduction: Traffic accidents imply congestion, delays, economic losses, disability persons and sometimes loss of human life. There are many factors influencing the likelihood of occurrence and severity. These factors include driver-related issues, topography and road-related issues, weather-related issues, and other accident-related issues. Predicting the severity of road accidents and understanding the factors that cause them are interesting research goals in traffic safety. This thesis analyses many traffic accidents deeply and determines the severity of accidents by using machine learning techniques. We also identified the important factors that have a direct impact on the severity of a traffic accident. This knowledge can help trainers to better educate new drivers to avoid traffic accident and can help policy maker in enforcing new laws that help reducing the number of severity of these accidents.
Methodology: Analysis has been done using Support Vector Machine (SVM), Artificial Neural Networks (ANN), and Random Forest (RF). Cross-validation with 10-fold was used to evaluate the performance. Before applying machine learning techniques, traffic accidents data was preprocessed through three stages: handling missing data, handling text and categorical attributes, and feature scaling.
Results: In this thesis, we had classified the severity of an accident into four classes based on the time delay after the accident. Our results, indicates that the most important factors that have a direct impact on the severity were time duration, end latitude, end longitude, start longitude, start latitude, and distance(mi).
Conclusion: Considering the overall accuracy, RF classifier was outperformed with (93.45%), followed by ANN (90.18%), and SVM (89.74%).
Keywords: Road accident, Support Vector Machine (SVM), Artificial Neural Networks (ANN), Random Forest(RF)