A Comparative Analysis of Data Imputation Methods for Missing Traffic Flow Data

To deal with the missing data problem in traffic flow datasets,a variety of missing data estimation meth-ods,including temporal correlation based methods,spatial correlation based methods,and spatial-temporal correlation based methods,are studied in this paper.The temporal correlation based methods include historical data based method, moving average method,exponential smoothing method,and linear regression method.The spatial correlation based method uses data collected from adjacent lanes and detectors to complete the missing data,while the spatial-temporal cor-relation based method considers both temporal and the spatial correlation of traffic flow.These methods are evaluated by actual traffic data collected from the freeway I-880 in California,USA.The results show that the method of exponential smoothing with smooth coefficient α=0.1,and the weighted average method based on the data of adjacent lanes outper-formed others.