Neuroimaging algorithms for GPS Trajectories Clustering

The mass use of portable devices such as smartphones and smartwatches allows us to have a complete history of our life. Thanks to the information collected by these terminals, we are able to reconstruct, with extreme accuracy, the history of our movements.

The application of machine learning algorithms to positioning data allows us to extract very relevant information useful for understanding our behavior. Some of these patterns are for example: the path we take from home to work, trips outside the city, which places we go to on Saturday nights etc ..

In this article we will describe an efficient and practical unsupervised machine learning algorithm to be able to carry out the identification of trajectory patterns. In the first part we will describe the clustering algorithm used to discover patterns, in the second one we will show how to implement this algorithm concretely in Python.

A Neuroimaging Algorithm for the Clustering of GPS Trajectories

Instead of using classic clustering algorithms like KMeans or DBSCAN, in this article we will use a cluster algorithm commonly used in brain image analysis.

QuickBundle (QB) (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3518823/pdf/fnins-06-00175.pdf) is a simple and compact clustering algorithm used to cluster matter fibers white brain extracted through the application of tractography algorithms (https://en.wikipedia.org/wiki/Tractography).

From a simple observation of the images shown below, we can easily note that GPS trajectories are very similar to white matter fibers.