A Transfer Learning–Based LSTM for Traffic Flow Prediction with Missing Data

Abstract

Traffic flow prediction plays an important role in intelligent transportation systems (ITS) on freeways. However, incomplete traffic information tends to be collected by traffic detectors, which is a major constraint for existing methods to get precise traffic predictions. To overcome this limitation, this study aims to propose and evaluate a new advanced model, named transfer learning–based long short-term memory (LSTM) model for traffic flow forecasting with incomplete traffic information, that adopts traffic information from similar locations for the target location to increase the data quality. More specifically, dynamic time warping (DTW) is used to evaluate the similarity between the source and target domains and then transfer the most similar data to the target domain to generate a hybrid complete training sample for LSTM to improve the prediction performance. To evaluate the effectiveness of the transfer learning–based LSTM, this study implements empirical studies with a real-world data set collected from a stretch of I-15 freeway in Utah. Experimental study results indicate that the transfer learning–based LSTM network could effectively predict the traffic flow conditions with a training sample with missing values.