Robust and flexible strategy for missing data imputation in intelligent transportation system

Rich and complete data play a fundamental role in intelligent traffic management and control applications. A great volume of missing data is found in the intelligent transportation system. In this paper, the authors introduce an ensemble strategy to handle the missing values. The proposed strategy is a general framework that different models, whether linear, neural networks, or other, can be applied. In this strategy, missing values are first computed by the forward and backward models, and their results are combined to recover the incomplete raw data. Then, the models are iterated for several times to enhance the accuracy. Three commonly used imputation models are tested in the proposed strategy using the data from real world. The results indicate that the proposed strategy can significantly improve the accuracy of the imputation with different missing types and during different traffic states. Moreover, the increase of the iteration is capable to improve the performance of the models.