Destination Prediction by Identifying and Clustering Prominent Features from Public Trajectory Datasets

Abstract

Destination prediction is an essential task in many location-based services (LBS) such as providing targeted advertisements and route recommendations. Most existing solutions were generative methods that model the problem as a series of probabilistic events that are then used to compute the destination probability using Bayes’ rule. In contrast, we propose a discriminative method that chooses the most prominent features found in a public trajectory dataset, clusters the trajectories into groups based on these features, and performs destination prediction queries accordingly. Our method is more concise and simple than existing methods while achieving better runtime efficiency and prediction accuracy as verified by experimental studies.