From traces to trajectories: How well can we guess activity locations from mobile phone traces?

Passively generated mobile phone dataset is emerging as a new data source for research in human mobility patterns. Information on individuals' trajectories is not directly available from such data; they must be inferred. Many questions remain in terms how well we can capture human mobility patterns from these datasets. Only one study has compared the results from a mobile phone dataset to those from the National Household Travel Survey (NHTS), though the comparison is on two different populations and samples. This study is a very first attempt that develops a procedure to generate a simulated mobile phone dataset containing the ground truth information. This procedure can be used by other researchers and practitioners who are interested in using mobile phone data and want to formally evaluate the effectiveness of an algorithm. To identify activity locations from mobile phone traces, we develop an ensemble of methods: a model-based clustering method to identify clusters, a logistic regression model to distinguish between activity and travel clusters, and a set of behavior-based algorithms to detect types of locations visited. We show that the distribution of the activity locations identified from the simulated mobile phone dataset resembles the ground truth better than the existing studies. For home locations, 70% and 97% of identified homes are within 100 and 1000 m from the truth, respectively. For work places, 65% and 86% of the identified work places are within 100 and 1000 m from the true ones, respectively. These results point to the possibility of using these passively generated mobile phone datasets to supplement or even replace household travel surveys in transportation planning in the future. (C) 2014 Elsevier Ltd. All rights reserved.