Parameter Estimation and Pattern Validation in Flock Mining

Abstract

Due to the diffusion of location-aware devices and location-based services, it is now possible to analyse the digital trajectories of human mobility through the use of mining algorithms. However, in most cases, these algorithms come with little support for the analyst to actually use them in real world applications. In particular, means for understanding how to choose the proper parameters are missing. This work improves the state-of-the-art of mobility data analysis by providing an experimental study on the use of data-driven parameter estimation measures for mining flock patterns along with a validation procedure to measure the quality of these extracted patterns. Experiments were conducted on two real world datasets, one dealing with pedestrian movements in a recreational park and the other with car movements in a coastal area. The study has shown promising results for estimating suitable values for parameters for flock patterns as well as defining meaningful quantitative measures for assessing the quality of extracted flock patterns. It has also provided a sound basis to envisage a formal framework for parameter evaluation and pattern validation in the near future, since the advent of more complex pattern algorithms will require the use of a larger number of parameters.