A study of periodic behaviors

Abstract

The last few years have witnessed an increasing popularity of smartphones with full GPS capability onboard, which results in the collection of large spatial-temporal datasets and the opportunity of discovering useful information on people movement behavior to foster innovative location services. Understanding period behaviors plays the key role in mining people movements. Nevertheless, sequential pattern mining

could be complicated, considering multiple periodic behaviors, interleaving periods, spatial-temporal noises and outliers.

In this project, we propose a novel approach to first detect periods in complex movements by finding reference locations with interpolated GPS data using kernel density estimation and their associated periods using Fourier transform and autocorrelation. At the second stage, a periodic mobility model is proposed to characterize the periodic behaviors. It is our belief that the model is directly applicable in mining periodic behaviors for moving users of telecommunication companies with

their GPS data or moving users of location-based social networks with their check-in data.