
Sadilek evaluated a large dataset consisting of 703 subjects (carrying GPS devices) over a variety of different time periods and collected more than 30,000 daily samples.
"While your location in the distant future is in general highly independent of your recent location, as we will see, it is likely to be a good predictor of your location exactly one week from now. Therefore, we view long-term prediction as a process that identifies strong motifs and regularities in subjects’ historical data, models their evolution over time, and estimates future locations by projecting the patterns into the future," Sadilek writes.
The system, called Far Out, could be used to map future traffic congestion, disease spread, and electricity demand.
No comments:
Post a Comment