The prediction of potential risk path in railway traffic events

In railway traffic operation, the prediction of risk path is one of the important issues because it can ensure the potential consequences are effectively mitigated and controlled to prevent the domino effect. However, it is quite difficult to mine the potential information and investigate the complex dependency in failure text data, which makes the prediction of potential risk path challenging. In this paper, we propose a new network-based risk prediction model to investigate the propagation path of potential risk and reduce the risk of cascade failures. Three kinds of information hidden in network connections are considered: local structural information, global structural information and attribute information. The model uses the keyword extraction method of text data for data preprocessing. The breadth-first search-based algorithm is improved to identify the meta-paths. The co-occurrence matrix and the association matrix are considered to play a role in the model. In order to verify the feasibility and advantages of the model, we use a dataset consisting of traffic events in Beijing subway as a case study. Results of the comparative analysis show that the proposed model not only can effectively predict the potential risk path, but also provides the best results in terms of ROC, AUC and Precision.