A Hybrid Finite Mixture Model For Exploring Heterogeneous Ordering Patterns Of Driver Injury Severity

With the accelerating process of motorization,more and more countries faced with an increasing traffic safety pressure.In academia,using the statistical model to analyze traffic accident data and exploring the complicated relation between accident factors and injury severity have become an important branch in the field of transportation.In this area,the characteristics of the model and the fitting effect directly affect the accuracy of accident data mining.Models without proper econometrical structures for accommodating the complex ordering patterns of injury severity could result in biased estimations and misinterpretations of factors.This paper proposes a hybrid finite mixture(HFM)model aiming to capture heterogeneous ordering patterns of driver injury severity while enhancing modeling flexibility.Firstly,by analyzing traffic accident factors,we select 17 important factors in driver;vehicle;road;environment aspects and divide injury severity into five categories according to KACBO classification.Then selecting part of the motor vehicle accident data of General Estimates System(GES)as sample.Second,considering the complexity of ordering patterns of injury severity,this paper proposes a HFM model which has two different multiple regression models.It attempts to probabilistically partition samples into two groups by EM algorithm inwhich one group represents an unordered/nominal data-generating process while the other representsan ordered data-generating process.Although some studies use traditional finite mixture model on the analysis of injury severity,the components of traditional finite mixture model are the models.Hence there are limitations on ordering patterns of injury severity.HFM model attempts to circumvent the distortion problem caused by single classification pattern(unordered or ordered)compared with traditional finite mixture model.It also creates a new field on finite mixture model.Finally,this paper compares the proposed HFM model with the other four models,including empirical results and analyses and three different types of evaluation criterias.The results prove that all HFM model and the other four models consider that more than 30 years old age,women drivers and rear-end collision will aggravate the degree of injury severity.However,vehicle trailing will alleviate degree of injury severity.In addition,unordered patterns of injury severity dominate the data-generating process in HFM model.Compared with the multiple regression model and the traditional finite mixture model,HFM model has a best comprehensive evaluation.It can effectively capture the heterogeneity of latent factors and dig more information in the data.