On the inclusion of spatial information for spatio-temporal neural networks

When confronting a spatio-temporal regression, it is sensible to feed the model with any available prior information about the spatial dimension. For example, it is common to define the architecture of neural networks based on spatial closeness, adjacency, or correlation. A common alternative, if spatial information is not available or is too costly to introduce it in the model, is to learn it as an extra step of the model. While the use of prior spatial knowledge, given or learned, might be beneficial, in this work we question this principle by comparing traditional forms of convolution-based neural networks for regression with their respective spatial agnostic versions. Our results show that the typical inclusion of prior spatial information is not really needed in most cases. In order to validate this counterintuitive result, we perform thorough experiments over ten different datasets related to sustainable mobility and air quality, substantiating our conclusions on real world problems with direct implications for public health and economy. By comparing the performance over these datasets between traditional and their respective agnostic models, we can confirm the statistical significance of our findings with a confidence of 95%.