Estimation of Travel Time from Taxi GPS Data

Traditionally travel time estimation is performed through data from loop detectors. However, this solution is not truly scalable because of the high cost associated with the installation and maintenance of loop detectors in large transportation networks. As GPS-equipped devices become increasingly common, it proves to be a more viable alternative data source for travel time estimation. Previous studies have successfully estimated travel time with good accuracy either from loop detectors data or GPS data. In this paper, we present a nearest-neighbor method for estimating travel time with partial information, using a distance measure derived from analytical models of the relationship between travel time and trip features. Our method is compared to a baseline nearest-neighbor method using generic Euclidean distance as its distance metric. We tested both methods on 1 million taxi trips and found that our method has successfully reduced the mean absolute percentage error (MAPE) value to 22.29% which is a 16% improvement over the baseline method.