Ensembling-mRBF-LSTM Framework for Prediction of Abnormal Traffic Flows

The prediction of abnormal traffic flows has always been a primary concern in traffic management. If the management unit can predict the occurrence of abnormalities, it can manage and control transportation in advance in order to avoid abnormal traffic flows and enhance the service quality. While previous researchers generally predicted abnormal conditions in traffic flows by means of a single time series prediction model, such methods might result in inaccuracy in the prediction of abnormal traffic flows in some areas. As a result, in terms of practicality, the traditional methods fail to produce satisfactory results. Moreover, with the traditional methods, researchers often find it difficult to obtain the best time-delay item in the model and can only do so by using the trial-and-error method, which is nevertheless cost-prohibitive. In order to solve this problem, in this study, we predicted abnormal traffic flows by using the Ensembling-mRBF-LSTM framework, which consists of three new concepts: (1) introducing the concept of ensembling learning to the target issue and integrating the results produced by multiple existing prediction methods into the prediction of abnormalities, (2) devising the concept of mRBF-LSTM to select the prediction methods applicable to the prediction of abnormalities in this study from the existing methods and to decide on the applicable time-delay item for each existing prediction method, and (3) predicting abnormalities by using only the existing methods selected via mRBF-LSTM. With the aforementioned three major concepts, in this study, we sought to vastly improve the existing method for abnormal traffic flow prediction. Lastly, this study provided the traffic flow data of Taipei Mass Rapid Transit (MRT) in Taiwan to verify the effectiveness of the proposed method.