Short-Term Traffic Prediction with Balanced Domain Adaptation

Short-term traffic forecasting has been a hot topic in the intelligent transportation systems field. The traditional traffic forecasting methods mostly fix traffic sensors. However, most sensors are subject to bad conditions, leading to noisy and insufficient raw data. Recent advances have provided new traffic prediction opportunities. For example, the transfer learning method takes advantage of data trained on one good dataset and transfers the knowledge to others with bad data. Existing applications do not consider the underlying data distributions sufficiently, limiting the prediction performance. We propose a transfer learning-based traffic flow prediction framework using the Balanced Domain Adaptation (BDA) method. Various regression models are fed into the framework to evaluate a good data source and predict bad target datasets. A case study using data from the Highways England is conducted. The results show that the proposed BDA-based framework can match the distributions between traffic flow datasets and significantly improve prediction accuracy.