Mixture Imputation Based Missing Value Imputing Method for Traffic Flow Volume Data

摘要

In order to improve the imputing performance of traffic missing data, it is vital important to analyze on imputing method and data pattern. Firstly, a mixture imputation( MI) based on iterated local least squares imputing (ILL Simpute) and Bayesian principal component analysis imputing( BPCAimpute) is introduced into handling the problem, which combines the global correlation of BPCAimpute and the local correlation of ILL-Simpute by distributing different weights. Secondly, a new kind of data pattern is used to analyze the traffic missing data, which can enhance the data correlation. Finally, the experiments prove that the MI provides a significantly better imputing performance than both BPCAimpute and ILLSimpute, the new data pattern also make a great contribution to a large extent.