RD-PCA: A Traffic Condition Data Imputation Method Based on Robust Distance

Abstract

Dynamic Transportation Information Service has penetrated into residents’ travels. The current problems that transportation information services face are variable such as real-time traffic forecasting, traffic managing and traffic induction. The above problems are related to the quality of historical traffic condition data. Due to a limited of GPS data collecting, the collected GPS data which scarcely covers the whole road network leads to incomplete and error traffic condition data. In consequence, two serious problems of traffic condition data quality manifest in incompleteness and low accuracy. This paper extends RD-PCA method which preliminarily focuses on the accuracy of imputing to prevent the estimating results from being impacted by outliers and aims at guaranteeing the completeness of imputing. The method excludes error data taking data quality measurement criterions. By adopting a measure factor, this method detects outliers and standardizes them, then constructs a robust feature space and imputes the missing data. The experimental results show that the proposed method can guarantee a high completeness and high accuracy under the condition of different missing rates.