Crash Severity Analysis of Highways Based on Multinomial Logistic Regression Model, Decision Tree Techniques, and Artificial Neural Network: A Modeling Comparison

The classification of vehicular crashes based on their severity is crucial since not all them have the same financial and injury values. In addition, avoiding by identifying influential factors possible via accurate prediction modeling. crash analysis, time-saving models are necessary for classifying severity. Moreover, statistical incapable potential regarding influencing incorporated in models. Unlike previous research efforts, which focused limited class severity, including property damage only (PDO), fatality, applying data mining models, present study sought to predict frequency according five levels PDO, severe injury, other visible injuries, complaint pain. multinomial logistic regression (MLR) model approaches, artificial neural network-multilayer perceptron (ANN-MLP) two decision tree techniques, (i.e., Chi-square automatic interaction detector (CHAID) C5.0) utilized traffic records State Highways California, USA. comparison findings relative importance ten qualitative quantitative independent variables CHAID C5.0 indicated that cause (X1) number vehicles (X5) were known as most involved crash. However, weather (X2) identified contributing ANN-MLP model. MLR showed driver’s age (X11) accounts a larger proportion Therefore, sensitivity analysis demonstrated had best performance predicting road Not did take shorter time (0.05 s) compared CHAID, MLP, MLR, it also represented highest accuracy rate training set. overall was approximately 88.09% 77.21% 70.21% MLP general, this revealed can be promising tool