Publication Details
Michiel Dhont, Elena Tsiporkova, Nicolás González-Deleito

2021 International Conference on Data Mining Workshops (ICDMW)

Contribution To Book Anthology


Mobility data typically covers both the spatial and temporal domain. The literature lacks methods which fully exploit the richness and at the same time manage the complexity of such data sets. Moreover, mobility data is often characterised by poor quality. In this paper two novel techniques are proposed that facilitate the advanced analysis of (spatio-temporal) mobility data. First, an incremental imputation technique is realised, which reduces substantially the amount of missing data by cleverly exploiting the spatio-temporal properties of historical data. Second, a multi-step non-negative matrix factorisation workflow is conceived allowing to extract spatio-temporal fingerprints of traffic trajectories of interest. The validation of both methods on a data set of vehicle counts from multiple adjacent locations in the Brussels-Capital Region (Belgium) produced already very promising results.