Predicting Road Class via Intrinsic Characterization by Leveraging on Fully Connected Layer and Topology.

Two primary approaches have been used for intrinsic road classification predic- tion. The more common method treats road characteristics as the dependent variable, using contextual information, such as Points of Interest and building data, as independent variables. However, the reliance on external contextual data introduces uncertainties regarding the applicability and scalability of these mod- els, particularly in regions where such data may be sparse or unavailable. A less explored method involves extracting information directly from road geometry, although previous attempts with this approach have resulted in low accuracy.

This daataset represent data used in our this paper, where we introduce a novel methodology that combines the principles of Persistent Homology (PH), a mathematical tool from computational topology that measures topological features across multiple scales, with a fully connected layer to predict road classes across diverse geographic contexts.