Classifying Location Points as Daily Activities using Simultaneously Optimized DBSCAN-TE Parameters.

Location-based services data collected from mobile phones represent a potentially powerful source of travel behavior data, but transforming the location points into semantic activities – where and when activities occurred – is non-trivial. Existing algorithms to label activities require multiple parameters calibrated to a particular dataset. In this research, we apply a simulated annealing optimization procedure to identify the values of four parameters used in a density-based spatial clustering with additional noise and time entropy (DBSCAN-TE) algorithm. We develop a spatial accuracy scoring function to use in the calibration methodology and identify paths for future research.