Taxis are an important part of the public transportation system in large cities, providing convenience for our daily life. In practice, the information contained in taxi trajectory data are imprecise and incomplete due to various factors such as measurement noise, low sampling rate, and geographic sparsity. In this thesis, we study the problem of data calibration and applications of knowledge discovery from taxi trajectory data, namely location information, passenger occupancy status, and travel speed. Specifically, we design an interactive map-matching system to involve human users in the loop to achieve high map-matching accuracy, and propose various query selection strategies to reduce the annotation cost effectively. Furthermore, we identify a new type of taxi fraud called u...[ Read more ]

Taxis are an important part of the public transportation system in large cities, providing convenience for our daily life. In practice, the information contained in taxi trajectory data are imprecise and incomplete due to various factors such as measurement noise, low sampling rate, and geographic sparsity. In this thesis, we study the problem of data calibration and applications of knowledge discovery from taxi trajectory data, namely location information, passenger occupancy status, and travel speed. Specifically, we design an interactive map-matching system to involve human users in the loop to achieve high map-matching accuracy, and propose various query selection strategies to reduce the annotation cost effectively. Furthermore, we identify a new type of taxi fraud called unmetered taxi rides, propose a learning model to predict the passenger occupancy status of taxis, and develop a heuristic algorithm to find fraudulent trajectories. Finally, we propose a learning model to predict the travel speeds of individual taxis, which is applied for detecting taxi speeding.