Developing a Crash Severity Prediction Model Using Machine Learning Technique
No Thumbnail Available
Date
2025-09
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Road Traffic crashes (RTCs) deaths and injuries are a universal global public health problem and significantly affect a country's economy, up to 3% of annual GDP. This paper examined the relationship between traffic flow features and expressway RTC severity. This study utilized a dataset containing detailed information on road traffic crashes and traffic flow. A comprehensive set of machine learning algorithms was involved, including LR, KNN, ADB, RF, XGB, NB, SVM, and DT, which were employed to develop the predictive model. The performance of the Model was assessed in terms of accuracy, precision, recall, and the F1 score. The best among the evaluated eight models for crash Severity prediction was the Random Forest model with an accuracy of 83%, a precision of 82.9 %, a recall of 83%, and an F1-score of 82.8 %. The study also found that chainage, months, crash types, and weather conditions are critical impacting features for crash severity prediction on the Addis-Adama Expressway. Moreover, traffic flow characteristics and Road traffic crash severity are weakly correlated. Thus, this study proves that the temporal and spatial factors are important for predicting RTC severity. The random forest model demonstrates to be an effective tool for forecasting RTC severity, which could be an effective model. Future work should focus on developing a prediction tool with an interactive map that could proactively monitor crash hot spots for potential crash severity in the Expressway, Ethiopia
Description
Keywords
RTC Analysis, Machine Learning Models, Crash Severity, Prediction Model, Confusion Matrix