Classification with Vertex-Based Graph Convolutional Neural Networks

Abstract

Graph convolutional neural networks (CNNs) use data that is supported on an arbitrary graph rather than a grid. Most current approaches to classification on graph data use spectral domain based graph CNNs and do not work for directed graphs. A more recent approach, topology adaptive graph convolution networks (TAGCN), uses graph based convolution from graph signal processing defined in the vertex domain instead of the spectral domain. In this paper, we use TAGCN to classify different time periods during the week based on New York City taxi data defined on a directed graph. We achieve an accuracy of 88% using a single graph convolutional layer. We classify the entire graph signal instead of classifying individual nodes in the graph.