Human movement is of interest to various disciplines:

• health (e.g. obesity, epidemiology)
• psychology (e.g. personality questionnaires: "I love to travel to places that I have never been before")
• urban planning (e.g. public transport)
• sociology/ economics (e.g. labour mobility)

Opportunity: smartphone location logs contain a wealth of easily accessible personal mobility data.

Challenge: the data is noisy, sparse and irregularly sampled. Missing data.

1. Build personalised map with discrete bins using entire history
2. Assign measurements to bins
3. Impute missing bins

We computed compared PMMI with other approaches (Barnett & Onnela 2016, Palmius et al. 2017).

• PMMI outperforms rival methods
• performance improves with the duration of the missing segment
• full coverage of missing periods
• ability to model uncertainty and complex non-linear interactions.

Further steps:

• Bayesian map building & measurement assignment
• More sophisticated classification methods

Barnett, I. and Onnela, J.-P. (2016). Inferring Mobility Measures from GPS Traces with Missing Data. arXiv:1606.06328 [stat].

Palmius, N., Tsanas, A., Saunders, K. E. A., Bilderbeck, A. C., Geddes, J. R., Goodwin, G. M., and Vos, M. D. (2017). Detecting Bipolar Depression From Geographic Location Data. IEEE Transactions on Biomedical Engineering, 64(8):1761-1771.