Mobile Computing, Applications, and Services. Second International ICST Conference, MobiCASE 2010, Santa Clara, CA, USA, October 25-28, 2010, Revised Selected Papers

Research Article

When Will You Be at the Office? Predicting Future Locations and Times

Download191 downloads
  • @INPROCEEDINGS{10.1007/978-3-642-29336-8_9,
        author={Ingrid Burbey and Thomas Martin},
        title={When Will You Be at the Office? Predicting Future Locations and Times},
        proceedings={Mobile Computing, Applications, and Services. Second International ICST Conference, MobiCASE 2010, Santa Clara, CA, USA, October 25-28, 2010, Revised Selected Papers},
        proceedings_a={MOBICASE},
        year={2012},
        month={10},
        keywords={location-prediction time-prediction},
        doi={10.1007/978-3-642-29336-8_9}
    }
    
  • Ingrid Burbey
    Thomas Martin
    Year: 2012
    When Will You Be at the Office? Predicting Future Locations and Times
    MOBICASE
    Springer
    DOI: 10.1007/978-3-642-29336-8_9
Ingrid Burbey1,*, Thomas Martin1,*
  • 1: Virginia Tech
*Contact email: iburbey@vt.edu, tlmartin@vt.edu

Abstract

The purpose of this paper is to predict people’s future locations or when they will be at given locations. These predictions support proactive, context-aware and social applications. Markov models have been shown to be effective predictors of someone’s location [1]. This paper incorporates temporal information in order to predict locations or the times when someone will be at a given location. Previous models use sequences of location symbols and apply Markov-based algorithms to predict the next location symbol. In our model, we embed temporal information within the sequence of location symbols. To predict a future location, we use the temporal information as the previous state (or context) in the Markov model to predict the location that is most likely at that given time. To predict someone will be at a location, we use the location as the context and predict the time(s) the person will be at that location. The model produces up to 91% accuracy for predicting locations, and less than 10% accuracy for predicting times. We show that prediction of location and prediction of time are two very different problems, because the number of predictions produced by the Markov model differ greatly between the two variables. A heuristic algorithm is proposed which incorporates additional context to improve predictions of future times to 43%.