Predictive Nearest Neighbor Queries over Uncertain Spatial-Temporal Data

Abstract

Predictive nearest neighbor queries over spatial-temporal data have received significant attention in many location-based services including intelligent transportation, ride sharing and advertising. Due to physical and resource limitations of data collection devices like RFID, sensors and GPS, data is collected only at discrete time instants. In-between these discrete time instants, the positions of the monitored moving objects are uncertain. In this paper, we exploit the filtering and refining framework to solve the predictive nearest neighbor queries over uncertain spatial-temporal data. Specifically, in the filter phase, our approach employs a semi-Markov process model that describes object mobility between space grids and prunes those objects that have zero probability to encounter the queried object. In the refining phase, we use a Markov chain model to describe the mobility of moving objects between space points and compute the nearest neighbor probability for each candidate object. We experimentally show that our approach can filter out most of the impossible objects and has a good predication performance.