A BPCA based missing value imputing method for traffic flow volume data

Abstract

Missing data problem widely exists in many traffic information systems, which brings great trouble to further studies. To solve this problem, this paper proposes a Bayesian principal component analysis (BPCA) based missing value imputing method to impute the incomplete traffic flow volume data collected in Beijing. Intuitively, this method takes an appropriate tradeoff between the historical and periodic information when imputing missing data. Experiments prove that the proposed method provides significant better imputing performance than two other frequently used imputing methods: historical imputing and spline imputing.