Abstract:
Due to the recent explosive growth of location-aware services, predicting the next position of a user is of increasing importance in enabling intelligent information services. To this effect, we first introduce a system where the Hidden Markov Model (HMM), a very popular tool used in fields like Natural Language Processing and Bioinformatics, is used to generate predictions over clustered location data. This method prioritizes the generality of the “area” of the user in favor of his or her exact co-ordinates. Secondly, frequent data reporting is expensive, and minimizing this whilst maintaining reasonable accuracy is preferable. On the flipside, security concerns like information leakage arise where hiding a user and preventing good prediction is preferable. Both these issues require us to analyze a relationship between the accuracy of future predictions against the frequency of current data being reported. This was our second end goal – we investigated how accuracy changed with data-reporting frequency (and other parameters) in an attempt to establish a model for accuracy variation.
