Spatial Autocorrelation Analysis with Graph Convolutional Neural Network

Spatial autocorrelation statistics have a long-standing history being used by geographers to

determine whether identifiable spatial patterns exist in data. However, existing research has

identified that solely relying on p-values can be problematic when working with large datasets.

This paper introduces a generalised model that can capture geographical data’s spatial

patterns using a graph convolutional network (GCN). The preliminary analysis demonstrates

that GCN can capture the localities among areas in local-scale datasets by processing the data

features and the spatial information separately into the graph network.