In order to quantitatively analyze the influence of highway dynamic congestion traffic information caused by traffic accidents and traveler heterogeneity on travel choice behavior, a large-scale SP/RP survey data covering 2 500 highway users in west Japan is used for this study. Latent class analysis(LCA) is adopted for extracting different preference features on traffic information, followed by a multilevel model which sets the results of LCA as explanatory variable. The research results show that：Through the potential category analysis of the data, travelers can be divided into three types of heterogeneous groups, namely, dynamic traffic information high dependence group, dynamic traffic information low dependence group and dynamic traffic information independent group, accounting for 38.8%, 36.1% and 25.1%. And there are significant differences in gender, age and occupation among these groups. The travel behaviors of travelers in different traffic information dependence groups are significantly different under dynamic traffic information. The travel prediction model considering data hierarchical structure is more accurate than the prediction model without data hierarchical structure.