Identification of population patterns using advanced machine learning techniques applied to mobile phone and geolocation data

Doctoral programme

Advanced Computing, Energy and Plasmas

Institution

University of Córdoba

Supervisor

Domingo Ortiz-Boyer

Summary

The proliferation of mobile devices and telecommunications infrastructure offers unprecedented opportunities to understand population mobility at fine spatial and temporal scales. This thesis develops and evaluates supervised and multi-label machine-learning methods for identifying meaningful places and mobility patterns from call detail records and geolocated social media. Through a thesis by compendium, the research first benchmarks supervised place identification against classical unsupervised clustering, then extends the task to mixed-use urban areas through multi-label learning and spatially aware neighbourhood selection. Case studies in Milan and Trento test performance across contrasting urban forms, while cross-border regions demonstrate how models learned from digital traces can transfer between geographic contexts. The resulting methods combine interpretable features, spatial validation and computationally efficient classifiers to improve the identification of home, work, density, forest and commuting patterns. Together, the studies provide a reproducible framework for mobility modelling, urban analytics and evidence-based planning while recognising the privacy, representativeness and transferability constraints of digital trace data.

Core publications