Recent research [MON2015] shows how deep learning methods (CNN) can achieve state-of-the-art classification performance on mobile phone metadata. These methods can exploit the temporal structure in mobile metadata by using specialized neural network architectures.


See the convnet-metadata repository on Github to learn how to use bandicoot weekmatrix features with the Caffe deep learning framework.

This module contains functions for outputting the week-matrix data representation, which can used with these deep learning methods. The mobile metadata is represented as 8 matrices summarizing mobile phone usage on a given week with hours of the day on the x-axis and the weekdays on the y-axis. These 8 matrices are the number of unique contacts, calls, texts and the total duration of calls for respectively incoming and outgoing interactions. Every cell in the matrices represents the amount of activity for a given variable of interest in that hour interval (e.g. between 2 and 3pm). In this way, any number of interactions during the week is binned. These 8 matrices are combined into a 3-dimensional matrix with a separate 'channel' for each of the 8 variables of interest. Such a 3-dimensional matrix is named a week-matrix.

create_weekmatrices(user[, split_interval]) Computes raw indicators (e.g.
read_csv(filename) Read a list of week-matrices from a CSV file.
to_csv(weekmatrices, filename[, digits]) Exports a list of week-matrices to a specified filename in the CSV format.


[MON2015]Felbo, B., Sundsøy, P., Pentland, A. S., Lehmann, S., & de Montjoye, Y. A. (2015). Using Deep Learning to Predict Demographics from Mobile Phone Metadata. arXiv preprint arXiv:1511.06660.