Classifying Roads with Multi-Step Graph Embeddings

Mohale E. Molefe, Jules R. Tapamo


Machine learning-based road-typeclassification is pivotal in intelligent road networksystems, where accurate network modelling is crucial.Graph embedding methods have emerged as theleading paradigm for capturing the intricate relationshipswithin road networks. However, their effectivenesshinges on the quality of input features. This paperintroduces a novel two-stage graph embeddingapproach used to classify road-type. The first stageemploys Deep Autoencoders to produce compactrepresentation of road segments. This compactifiedrepresentation is then used, in the second stage, bygraph embedding methods to generate an embeddedvectors, leveraging the features of neighbouringsegments. Results achieved, with experiments onrealistic city road network datasets, show that theproposed method outperforms existing approaches withrespect to classification accuracy.


Road type classification, road networks intelligent systems, graph embedding methods, deep autoencoder

Full Text: PDF