Machine learning-based classification of roads is crucial for the effectiveness of smart transportation infrastructure. Recent road-type classification methods rely on road attribute features obtained from the publicly available OpenStreetMaps platform. However, the absence of certain road attributes often makes it difficult to develop models for monitoring road networks using OpenStreetMaps. Presented in this study is an approach for classifying roads where features are extracted using node structural feature generation methods and Graph Convolution Networks; without relying on road attribute features. The proposed method was evaluated on road network graph datasets of realistic cities. Experiments demonstrate that the presented method is capable of effectively modelling road networks even with missing road attributes. The results indicate that the proposed method can model road networks with missing road attributes.