Detecting Outliers in Cell Phone Data

Abstract

The use of cell phone signaling data for traffic modeling has great potential. Because of the high coverage rate of these phones, the data can be used as an addition to or even as a replacement for stationary sensors when the deployment of stationary sensors is not possible or too expensive. However, cell phone signaling data are error-prone and have to be preprocessed for use in traffic modeling. First, the positions reported by cell phone signaling data may be inaccurate. Second, because of privacy issues, additional data may be introduced to obfuscate actual movements. This study presents three filters to rectify the trajectories generated by cell phone movements. For evaluation, the filters were applied to cell phone trajectories and compared with corresponding GPS-based tracks. The evaluation data covered 4,933 tracks collected automatically and 5 tracks collected manually. The proposed filters significantly improved estimation of speed and position compared with the raw trajectories of cell phone movements.