Traffic Crash Severity: Comparing the Predictive Performance of Popular Statistical and Machine Learning Models Using the Glasgow Coma Scale

Nazir, M and Illahi, U and Gurjar, J and Mir, MS (2023) Traffic Crash Severity: Comparing the Predictive Performance of Popular Statistical and Machine Learning Models Using the Glasgow Coma Scale. In: Journal of The Institution of Engineers (India): Series A .

PDF jou_ins_eng_2023.pdf - Published Version Restricted to Registered users only Download (801kB) | Request a copy

Abstract

Crash severity analysis and prediction is a promising field in traffic safety. Various statistical methods have been used to model the severity of road crashes. However, machine learning algorithms have gained popularity in recent years. This study compares the predictive performance of various machine learning and statistical models, including prediction accuracy, and determines the influence of various variables on crash severity. The crash severity data were collected from a Hospital in Kashmir (India), an area with mixed topography. The crash severity levels (CSLs) were represented in the Glasgow Coma Scale (GCS). For estimations, the two statistical models, logistic regression (LR) and decision tree (DT), and four machine learning models, including random forest (RF), support vector machine (SVM), gradient boosted tree (GBT), and extreme gradient boosting (XG BOOST), have been used. The results show that the machine learning models have higher prediction accuracy than the statistical models. Among all, the GBT model has the best overall prediction accuracy, particularly in the prediction of individual CSLs while LR was found to have the least accuracy. The influence of variables on CSL was found from DT and GBT. Both models have indicated that ‘time’ as a variable was the most influencing, followed by the casualty class of pedestrians over the CSLs. The results also show that the variable influences over CSL were different from different models. Based on the influence of variables, certain policy implications are suggested, which might aid the transportation department, and other concerned departments to reduce the severity and number of road traffic crashes (RTCs). © 2023, The Institution of Engineers (India).

Item Type:	Journal Article
Publication:	Journal of The Institution of Engineers (India): Series A
Publisher:	Springer
Additional Information:	The copyright for this article belongs to Springer.
Keywords:	Adaptive boosting; Forecasting; Learning systems; Logistic regression; Motor transportation; Public policy; Roads and streets; Statistics; Support vector machines; Topography, Crash severity; Glasgow coma scale; Logistics regressions; Machine learning models; Machine-learning; Policy implications; Prediction accuracy; Predictive performance; Severity prediction; Statistic modeling, Decision trees
Department/Centre:	Division of Mechanical Sciences > Civil Engineering
Date Deposited:	15 Mar 2023 06:04
Last Modified:	15 Mar 2023 06:04
URI:	https://eprints.iisc.ac.in/id/eprint/80989

Actions (login required)

View Item