Predicting future locations using prediction-by-partial-match
Ingrid Burbey, Thomas L. Martin · 2008
We implemented the Prediction-by-Partial-Match data compression algorithm as a predictor of future locations. Positioning was done using IEEE 802.11 wireless access logs. Several experiments were run to determine how to divide the data for training and testing and how to best represent the data as a string of symbols. Our test data consisted of 198 datasets containing over 28,000 pairs, obtained from the UCSD Wireless Topology Discovery project. Tests of a first-order PPM model revealed a 90% success rate in predicting a user's location given the time. The third-order model, which is given the previous time and location and asked to predict the location at a given time, is correct 92% of the time.