The rapid development of technology and the increased use of information systems has led to the creation and collection of spatio-temporal data such as traffic, crime and seismic activity data. A characteristic of this type of data is that it includes patterns that are difficult to identify with conventional ways of analysis, in order to draw conclusions and predictions. On the contrary, it has been found that artificial intelligence tools, specifically deep learning, show satisfactory results, in terms of their ability to perceive high-level correlations in spatio-temporal historical data. In cases of prediction where the complexity and amount of primary spatio-temporal data is great, such as for example spatio-temporal crime prediction, traffic prediction, pandemic evolution prediction, and even human activity recognition, deep learning models show satisfactory accuracy in their results. Nevertheless, the effectiveness of these models depends, among other things, to a significant extent on their architecture as well as the structure of the data with which they are trained. Based on the above, the object of the proposed research is the development, testing and evaluation of deep learning algorithms on different sets of spatiotemporal data in order to create innovative tools for predicting spatiotemporal values. This will be achieved through:

(a) Critical overview of the current state of the art in applying deep learning techniques to spatiotemporal datasets.

(b) The design and development of new spatiotemporal deep learning tools for spatiotemporal prediction.

(c) Testing and evaluating the tools that will be developed.