VOut: Mining Sampled GPS Trace for Taximeter Fraud

Abstract

Nowadays Taxies play an important role in urban transportation. One practical issue in taxi operation is charging fraud, where the taximeter of a malicious taxi is altered to fake the real travel distance. One labor-intensive scheme to detect fraudulent taxies is to check the taximeters of a great number of randomly selected taxies or those taxies been accused of cheating. Alternatively, with GPS (Global Positioning System) receiver onboard, another potential scheme is to calculate the travel distance of a taxi according to the GPS reports and a digital map. However, it is hard to retrieve accurate trajectories of taxies due to the sparseness of GPS reports, outlier reports, and the distortion of the digital map. In this paper, we propose a speed-based scheme, called VOut, which relies on one key insight that the average speed derived from a faked travel distance should be higher than the majority of GPS velocity reports. Given the operation records and GPS reports of taxies, VOut judges the suspicion of each delivery task for each taxi. We conduct extensive trace-driven simulations and the results demonstrate the efficacy of VOut.