Abstract
Abstract. The number of GPS trajectories recorded daily has been continuously growing in the recent years and new methods to analyse such big data are surfacing all the time. In this paper, we focus on destination prediction, which is useful in various applications like hazard detection and advertisement. We proposed a real-time method for destination prediction of moving users. It uses the current movement trajectory of the user together with historical and regional information to make an accurate prediction. The method is efficient because we can rapidly compute features with the help of spatial and non-spatial indexing methods. We tested the method with real trajectories collected by Mopsi users. The success rate of the method is up to 65 % depending on the length of the recorded trajectory so far, i.e. how long the user has been on move. To our knowledge, this is the first real-time system capable of such success.
Highlights
Location-based services are becoming more and more popular, and the market size is said to reach 68.85 billion US dollars by 2023 (Marketsandmarkets, 2018)
One source for the potential destinations is the Mopsi services database2, which is a collection of Points of interest (POIs) such as restaurants, shops and hotels among many others
The dataset contains the most recent trajectories from 10 different Mopsi users recorded by 31.3.2019. These contain a total of 2,484 trajectories consisting of 3,409,812 points, travelling of a total 30,599 km in 2,103 hours
Summary
Location-based services are becoming more and more popular, and the market size is said to reach 68.85 billion US dollars by 2023 (Marketsandmarkets, 2018). In this work we devised a method to guess the end-point using information about the current movement, the surroundings, and user history This goal is not easy to achieve mostly because trajectories have varying lengths. In (Krumm, 2010; Krumm, 2016) prediction is limited to only the very near future These are probability-based methods that use Markov models to determine what happens at the following intersection: does the user turn or continue forward. While these methods work with good accuracy, using them for long-term prediction has the side effect of small errors propagating and leading to a much lower accuracy. An alternative use is in advertising where on-path services are retrieved using predictive range queries (Jeung et al, 2010) on spatial databases
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
