GDGraphNet: A Geodemographic Graph Neural Network for Sociodemographics

Geodemographics studies populations based on where people live by combining demographic characteristics, such as age, income, and education, with geographic information to support policymaking. Since geodemographic data can naturally be represented as graphs, Graph Neural Networks (GNNs) provide a promising framework for this task. However, their application in geodemographics remains underexplored. To address this gap, we propose the Geodemographic Graph Neural Network (GDGraphNet), a lightweight but effective framework consisting of a Multi-Layer Perceptron encoder for capturing demographic patterns, a GNN decoder for incorporating spatial information, and a Correlation Learner for preserving feature correlation distributions. We evaluate GDGraphNet on the Output Area Classification (OAC) task using the England and Wales Census 2021 dataset with geographic information, comprising 181,408 nodes with 61 features each. Experimental results demonstrate that GDGraphNet consistently outperforms the official OAC results and several popular GNN-based methods across the Silhouette, Calinski–Harabasz, and Davies–Bouldin metrics. The source code used in this work is publicly available at: https://github.com/cat-lovesdonuts/GDGraphNet.