• Deep learning
• Human mobility
• Machine learning
• Crowd flow prediction

The last years have witnessed a significant growth of human mobility studies, motivated by their importance in a wide range of applications, from traffic management to public security, computational epidemiology and pollution monitoring.
Among the many tasks involving mobility data, we focus on crowd flow prediction, i.e., forecasting incoming and outgoing flows in the locations of a geographic region.
Although several deep learning approaches have been proposed to solve this problem, their usage is limited to specific types of tessellations and cannot provide sufficient explanations of their predictions.
In this thesis, we propose AdjNet (Adjacency Matrix Neural Network), which solves crowd flow prediction using an approach based on Graph Convolutional Networks (GCN).
We use a public dataset on crowd flows described by bike trips to compare the predictive performance of AdjNet with that of STResNet, a state-of-the-art model for crowd flow prediction.
Although the two approaches have comparable performance, AdjNet presents several improvements with respect to STResNet.
First, AdjNet can be used with regions of irregular shapes and can provide information about the origin and destination of the predicted crowd flows.
Second, AdjNet can provide meaningful explanations of the predicted in- and out-crowd flows.
Our model is a step forward in the design of reliable deep learning models to predict and explain human displacements in urban environments.