Panda ∗: A generic and scalable framework for predictive spatio-temporal queries

Abstract

Predictive spatio-temporal queries are crucial in many applications. Traffic management is an example application, where predictive spatial queries are issued to anticipate jammed areas in advance. Also, location-aware advertising is another example application that targets customers expected to be in the vicinity of a shopping mall in the near future. In this paper, we introduce Panda?, a generic framework for supporting spatial predictive queries over moving objects in Euclidean spaces. Panda? distinguishes itself from previous work in spatial predictive query processing by the following features: (1) Panda? is generic in terms of supporting commonly-used types of queries, (e.g., predictive range, KNN, aggregate queries) over stationary points of interests as well as moving objects. (2) Panda? employees a prediction function that provides accurate prediction even under the absence or the scarcity of the objects' historical trajectories. (3) Panda? is customizable in the sense that it isolates the prediction calculation from query processing. Hence, it enables the injection and integration of user defined prediction functions within its query processing framework. (4) Panda? deals with uncertainties and variabilities in the expected travel time from source to destination in response to incomplete information and/or dynamic changes in the underlying Euclidean space. (5) Panda? provides a controllable parameter that trades low latency responses for computational resources. Experimental analysis proves the scalability of Panda? in evaluating a massive volume of predictive queries over large numbers of moving objects.