Parameter Optimisation for Location Extraction and Prediction Applications

The pervasive nature of location-aware hardware has provided an unprecedented foundation for understanding human behaviour. With a record of historic movement, in the form of geospatial trajectories, extracting locations meaningful to users is commonly performed as a basis for modelling a users´ interactions with their environment. Existing literature, however, has scarcely considered the applicability of extracted locations, typically focusing solely on the consequent knowledge acquisition process employed, due to the difficulty of evaluating the output of such unsupervised learning techniques. Towards the goal of ensuring the representativeness of extracted locations, and using location prediction as an example knowledge acquisition process, this work provides a method of automated parameter selection for both location extraction and prediction that ensures both the applicability of the locations extracted and the utility of the predictions performed. Specifically, we: (i) provide a metric for the evaluation of both extracted locations and predictions that characterises the goal of each of these tasks, (ii) frame the process of parameter selection as that of mathematical optimisation through the presented metric, and (iii) discuss characteristics of the metric while demonstrating its applicability over real-world data, location extraction algorithms and prediction techniques.