DepGCN: A Spatial Deep Learning Framework Based on Graph Convolutional Networks for Modeling Depression Prevalence
Loading...
Authors
Karami, Fahime
Date
2025-05-29
Type
thesis
Language
eng
Keyword
Deep learning , GIS , Depression modeling
Alternative Title
Abstract
Depression is a common mental health disorder that affects millions of people around the world and has a substantial impact on individual, societal and economic outcomes. Even with this commonality, the great complexity of modeling depression — influenced by numerous socioeconomic, environmental, and community factors — still presents a challenge for accurate modeling. Traditional models are poor at capturing fine-grained spatial dependencies and localized variations, which can be important for understanding depression prevalence. Spatial analysis, which accounts for geographic relationships and neighborhood effects, is a promising approach for revealing these patterns, but the combination of spatial analysis with advanced machine learning methods, and especially deep learning, has not yet been sufficiently explored.
To address this gap, we proposed a deep learning framework called DepGCN to model the prevalence of depression by combining Graph Convolutional Networks (GCNs) and Neural Networks (NNs). The GCN block captures spatial correlations and neighborhood effects by aggregating information of spatial adjacent regions, and the NN block is to refine the features extracted by the GCN block to create an accurate model. DepGCN combines the power of spatial analysis and deep learning, allowing the model to discover latent patterns in data that classical approaches often overlook.
This study is focused on New York State and uses census tract-level data, including 121 variables classified into ten themes, including demographic characteristics, socioeconomic indicators, housing stability, environmental quality, and access to healthcare services. These variables were carefully chosen to include different elements leading to depression, constituting a comprehensive basis for the model. suited for localized public health interventions.
The proposed approach, DepGCN, outperforms traditional methods. Compared to the best baseline, called Support Vector Regression, DepGCN enables a 5.68\% improvement in $\mathrm{R^2}$ and a 32\% reduction in the mean squared error. The findings emphasize the importance of location-based research in mental health modelling, as well as the promise of deep learning in public health research. By providing a new perspective on the prevalence of depression, DepGCN can help to inform interventions based on localized hotspots, thus offering actionable intelligence to policymakers and public health officials.
Description
Citation
Publisher
License
Queen's University's Thesis/Dissertation Non-Exclusive License for Deposit to QSpace and Library and Archives Canada
Intellectual Property Guidelines at Queen's University
Copying and Preserving Your Thesis
This publication is made available by the authority of the copyright owner solely for the purpose of private study and research and may not be copied or reproduced except as permitted by the copyright laws without written authority from the copyright owner.
Attribution-NonCommercial-NoDerivatives 4.0 International
Intellectual Property Guidelines at Queen's University
Copying and Preserving Your Thesis
This publication is made available by the authority of the copyright owner solely for the purpose of private study and research and may not be copied or reproduced except as permitted by the copyright laws without written authority from the copyright owner.
Attribution-NonCommercial-NoDerivatives 4.0 International