Crash Injury Severity Prediction Using an Ordinal Classification Machine Learning Approach

In many related works, nominal classification algorithms ignore the order between injury severity levels and make sub-optimal predictions. Existing ordinal classification methods suffer rank inconsistency and rank non-monotonicity. The aim of this paper is to propose an ordinal classification approach to predict traffic crash injury severity and to test its performance over existing machine learning classification methods. First, we compare the performance of the neural network, XGBoost, and SVM classifiers in injury severity prediction. Second, we utilize a severity category-combination method with oversampling to relieve the class-imbalance problem prevalent in crash data. Third, we take advantage of probability calibration and the optimal probability threshold moving to improve the prediction ability of ordinal classification. The proposed approach can satisfy the rank consistency and rank monotonicity requirement and is proved to be superior to other ordinal classification methods and nominal classification machine learning by statistical significance test. Important factors relating to injury severity are selected based on their permutation feature importance scores. We find that converting severity levels into three classes, minor injury, moderate injury, and serious injury, can substantially improve the prediction precision.

Analyzing Accident Injury Severity via an Extreme Gradient Boosting (XGBoost) Model

Vehicle to vulnerable road user (VRU) crashes occupy a large proportion of traffic crashes in China, and crash injury severity analysis can support traffic managers to understand the implicit rules behind the crashes. Therefore, 554 VRUs-involved crashes are collected from January, 2017, to February, 2021, in a city in northern China, including 322 vehicle-pedestrian crashes and 232 vehicle-bicycle crashes. First, a descriptive statistical analysis is conducted to investigate the characteristics of VRUs-involved crashes. Second, the extreme gradient boosting (XGBoost) model is introduced to identify the importance of risk factors (i.e., time of day, day of week, rushing hour, crash position, weather, and crash involvements) of VRUs-involved crashes. The statistical analysis demonstrates that the risk factors are closely related to VRUs-involved crash injury severity. Moreover, the results of XGBoost reveal that time of day has the greatest impact on VRUs-involved crashes, and crash position shows the minimum importance among these risk factors.

Exploring Factors Contributing to Crash Injury Severity in the Rural-Urban Fringe of the Central City

To analyze the risk factors influencing the crash injury severity in rural-urban fringes, crash data in rural-urban fringes were collected from Harbin, China. Four risk factors, namely, time of day, vehicle type, road feature, and crash type, were investigated associated with the severity of rural-urban fringe crashes. The crash injury severity was divided into two categories, including fatal and nonfatal crash. The logistic regression was applied to explore the relationships between the severity outcomes and time of day, vehicle type, road feature, and crash type. The test methods of goodness-of-fit and badness-of-fit are conducted to examine the validity of estimation results. The results show considerable matching of the number of different crash types between calculated results and actual data. Compared with the other influencing factors, the time of day is significant factor for crash injury severity based on the study. As such, the proposed calibration procedure and the factors of choice are recommended as a validated approach to analyze and identify the main factors influencing crash injury severity in rural-urban fringes.

Insights on Crash Injury Severity Control from Novice and Experienced Drivers: A Bivariate Random-Effects Probit Analysis

This study intended to investigate the crash injury severity from the insights of the novice and experienced drivers. To achieve this objective, a bivariate panel data probit model was initially proposed to account for the correlation between both time-specific and individual-specific error terms. The geocrash data of Las Vegas metropolitan area from 2014 to 2017 were collected. In order to estimate two (seemingly unrelated) nonlinear processes and to control for interrelations between the unobservables, the bivariate random-effects probit model was built up, in which injury severity levels of novice and experienced drivers were addressed by bivariate (seemingly unrelated) probit simultaneously, and the interrelations between the unobservables (i.e., heterogeneity issue) were accommodated by bivariate random-effects model. Results revealed that crash types, vehicle types of minor responsibility, pedestrians, and motorcyclists were potentially significant factors of injury severity for novice drivers, while crash types, driver condition of minor responsibility, first harm, and highway factor were significant for experienced drivers. The findings provide useful insights for practitioners to improve traffic safety levels of novice and experienced drivers.