A Study on Efficient Traffic Flow Data Cleaning Approaches

Enough and accurate traffic flow data was essential guarantee to realize Intelligent Transportation Systems.Many quality problems were existed inevitably in detected data, including inefficacy, redundancy, error, missing, time dot excursion etc.On the basis of sufficient study and analysis for the influence reasons of data quality, the definition of data cleaning was proposed, and the cleaning rules and cleaning steps of "dirty data" were studied at the same time.Then the proposed cleaning rules were calibrated with the detected data of loop vehicle detector.It is pointed that, the recognition rates of "dirty data" is up to 90%.The results show that, "dirty data" can be effectively detected to help to increase the validity and veracity of the following data mining according to cleaning rules and cleaning steps.