• data science
• generative models
• human mobility
• social network
• synthetic trajectories

The analysis of human mobility is crucial in several areas, from urban planning to epidemic modeling, estimation of migratory flows and traffic forecasting.
However, mobility data (e.g., Call Detail Records and GPS traces from vehicles or smartphones) are sensitive since it is possible to infer personal information even from anonymized datasets.
A solution to dealing with this privacy issue is to use synthetic and realistic trajectories generated by proper generative models.
Existing mechanistic generative models usually consider the spatial and temporal dimensions only.
In this thesis, we select as a baseline model GeoSim, which considers the social dimension together with spatial and temporal dimensions during the generation of the synthetic trajectories.
Our contribution in the field of the human mobility consists of including, incrementally, three mobility mechanisms, specifically the introduction of the distance and the use of a gravity-model in the location selection phase, finally, we include a diary generator, an algorithm capable to capture the tendency of humans to follow or break their routine, improving the modeling capability of the GeoSim model.
We show that the three implemented models, obtained from GeoSim with the introduction of the mobility mechanisms, can reproduce the statistical proprieties of real trajectories, in all the three dimensions, more accurately than GeoSim.