Enhancing location prediction with big data: evidence from dhaka

In recent years, the study of location prediction has received heightened attention due to its applications in LBS and other areas. However, most of the techniques and subsequent conclusions drawn from previous research works are specific to the data used in the study. For instance, resolution of location data and inclusion of external data (e.g., from social networks) may limit application of previous techniques to new situations. Therefore, we explore ways of enhancing location prediction techniques which leverage big data without the need for external data sources. To this end, we study a large CDR dataset with more than 3.5 billion calls from a leading cellular network provider in Dhaka, Bangladesh. The research question we tackle is how we can leverage big data to enhance performance of location predictors? Based on spatio-temporal analysis of call activity, we devise a scheme to compute prior probabilities from cell call activity. With this reasoning, we develop an enhanced Bayes predictor which uses a distance threshold and the users' regular location to improve generation of prior probabilities. Experimental results show that overall the enhanced Bayes predictor improves accuracy by 17 percentage points.