Subgraph Filtering for Fair Graph Neural Networks

Haohui Lu; jiyuan Tian; Fangyu Zhou; Shahadat Uddin

cs.LGAug 26, 2026

Subgraph Filtering for Fair Graph Neural Networks

Authors: Haohui Lu, jiyuan Tian, Fangyu Zhou, Shahadat Uddin

Organizations: Molly Wardaguga Institute for First Nations Birth Rights, Charles Darwin University, Darwin, Australia · Discipline of Business Analytics, The University of Sydney, Sydney, Australia · School of Electrical and Computer Engineering, The University of Sydney, Sydney, Australia

Abstract

Graph neural networks (GNNs) can exhibit unfair behavior even when sensitive attributes are excluded from node features, because graph topology and message passing propagate group-correlated signals under sensitive homophily. Existing fairness-aware GNN methods mainly constrain representations or prediction distributions at a global level, without explicitly controlling the local structural pathways through which biased information propagates during aggregation. We propose Subgraph Filtering for Fair Graph Neural Networks (SF-GNN), a lightweight and architecture-agnostic framework that mitigates structural bias at its source. SF-GNN identifies bias-prone edges by combining sensitive homophily with structural propagation amplifiers, including hub participation and triadic closure. It then incorporates stochastic edge filtering into each message-passing step to selectively downweight or remove these edges while preserving the remaining graph structure. Training further incorporates a statistical-parity regularizer with a warm-up schedule to stabilize optimization. Experiments on five benchmark datasets show that SF-GNN achieves consistent fairness improvements while maintaining competitive predictive performance, leading to a better fairness--accuracy trade-off than recent fairness-aware GNN baselines.

Figures & tables

Figure 1: Overview of the SF-GNN framework.

Method	NBA			German			Credit			Income			Pokec_n
	ACC $\uparrow$	$\Delta$ SP $\downarrow$	$\Delta$ EO $\downarrow$	ACC $\uparrow$	$\Delta$ SP $\downarrow$	$\Delta$ EO $\downarrow$	ACC $\uparrow$	$\Delta$ SP $\downarrow$	$\Delta$ EO $\downarrow$	ACC $\uparrow$	$\Delta$ SP $\downarrow$	$\Delta$ EO $\downarrow$	ACC $\uparrow$	$\Delta$ SP $\downarrow$	$\Delta$ EO $\downarrow$
GCN	62.53	9.50	13.86	65.12	6.45	4.22	66.03	3.45	6.42	63.83	14.68	13.00	84.91	1.56	2.82
GAT	65.57	10.43	11.00	65.12	7.78	7.62	67.29	3.47	7.06	62.46	16.34	13.29	70.75	4.45	4.44
GraphTrans	69.87	13.42	20.30	70.08	7.29	6.15	70.10	10.27	15.51	77.45	22.49	12.18	89.67	1.29	2.17
FairGNN	63.29	10.44	8.89	61.12	6.67	10.47	69.86	0.81	1.87	45.56	2.65	1.92	85.15	1.69	2.07
NIFTY	68.61	11.02	17.37	64.56	4.95	2.47	66.91	0.69	2.41	58.81	4.24	3.89	81.39	2.34	2.22

Table 1: Performance comparison across five datasets. ACC (

\uparrow

) indicates classification accuracy, while lower

\Delta

SP (

\downarrow

) and

\Delta

EO (

\downarrow

) indicate fairer outcomes. Results are averaged over 10 runs. Best results are shown in bold .

Dataset	Metric	Filtering Mechanism				Bias-Prone Edge Definition				Training / Control
Dataset	Metric	SF-GNN	No fairness	No SF	Drop	Downweight	HO	HH	HT	No FWU	RBPM	RFSNE
NBA	ACC $\uparrow$	72.15	70.38	72.13	70.67	72.06	72.05	72.01	71.05	72.06	70.13	72.01
	$\Delta$ SP $\downarrow$	2.29	6.54	9.20	8.63	6.61	4.44	8.80	3.84	14.05	5.89	7.43
	$\Delta$ EO $\downarrow$	9.10	14.02	19.44	10.76	11.81	14.18	13.63	14.22	11.30	19.17	9.17
German	ACC $\uparrow$	70.56	71.28	70.24	70.32	70.08	70.32	69.52	70.24	70.24	69.60	70.40
	$\Delta$ SP $\downarrow$	3.12	4.47	1.74	1.74	0.68	1.60	2.07	2.30	1.71	1.80	2.12

Table 2: Ablation study of Subgraph Filtering Graph Neural Network (SF-GNN) components across datasets. SF-GNN denotes the full proposed model. No fairness removes fairness regularization. No SF removes structure filtering. Drop and Downweight apply only edge dropping or edge downweighting. HO uses homophily-only edge identification, HH uses homophily + hub participation, and HT uses homophily + triangle closure. No FWU disables fairness warm-up, RBPM recomputes the bias-prone mask each epoch, and RFSNE randomly filters the same number of edges. We report ACC (

\uparrow

\Delta

SP (

\downarrow

), and

\Delta

EO (

\downarrow

). Best results for each metric are highlighted in bold.

Dataset	Metric	$p$			$q$			$\gamma$			$\lambda$
Dataset	Metric	Low	Mid	High	Low	Mid	High	Low	Mid	High	Low	Mid	High
NBA	ACC $\uparrow$	71.90	70.13	71.65	71.14	71.65	71.90	71.14	70.38	71.66	72.15	70.89	71.39
	$\Delta$ SP $\downarrow$	4.86	5.03	8.45	6.89	7.16	5.32	5.15	7.46	4.84	2.29	6.44	9.25
	$\Delta$ EO $\downarrow$	7.66	8.11	18.68	19.07	12.91	9.66	10.49	10.10	11.62	9.10	8.54	8.76
German	ACC $\uparrow$	70.40	70.40	70.16	70.00	69.44	70.40	70.00	70.16	70.40	70.56	70.16	69.68
	$\Delta$ SP $\downarrow$	0.63	1.99	2.90	1.57	1.68	1.53	2.06	1.66	2.27	3.12	1.42	3.52

Table 3: Sensitivity analysis of SF-GNN hyperparameters across datasets, evaluating the fairness regularization coefficient (

\lambda

), edge drop probability (

p

), downweight factor (

\gamma

), and hub quantile (

q

). Hyperparameters are tuned via grid search over

\lambda\in{0.5,0.6,0.7,0.8}

p\in{0.1,0.2,0.3}

\gamma\in{0.7,0.8,0.9}

, and

q\in{0.4,0.5,0.6}

. For each parameter, Low/Mid/High correspond to the smallest, middle, and largest tested values. We report ACC (

\uparrow

\Delta

SP (

\downarrow

), and

\Delta

EO (

\downarrow

Explore similar work

Geometrical fairness in graph neural networks
Jun 16, 2026Arturo Pérez-Peralta, Sandra Benítez-Peña, Blas Kolic +1Graph Neural Networks Graph Representation Learning
Department of Statistics, University Carlos III of Madrid, Spain · uc3m-Santander Big Data Institute
FairGC: Fairness-aware Graph Condensation
Mar 30, 2026Yihan Gao, Chenxi Huang, Wen Shi +5Algorithmic Fairness Group Fairness
College of Computer Science and Technology, Jilin University, Changchun, China · Key Laboratory of Advanced Design and Intelligent Computing, Dalian University, Dalian, China · School of Computing Technologies, RMIT University, Melbourne, Australia +1
Investigating the Effect of k-NN Preprocessing on Developing Graph Neural Networks: A Fairness-Based Perspective
Sep 27, 2026Nikolaos Zafeiropoulos, Emmanouil Mavrikos, George E. TsekourasGraph Neural Networks Algorithmic Fairness
Department of Cultural Technology and Communication University of the Aegean Mytilene, Greece

Jun 16, 2026J/K move · Enter open · S save

Geometrical fairness in graph neural networks

Arturo Pérez-Peralta, Sandra Benítez-Peña, Blas Kolic, Rosa E. Lillo

Department of Statistics, University Carlos III of Madrid, Spain · uc3m-Santander Big Data Institute

Graph Neural Networks Graph Representation Learning Algorithmic Fairness Spectral Methods Graph Diffusion

Open paper AskPDF arXiv

Graph-based learning methods have become increasingly prominent due to their strong performance across diverse applications. Among these, recent frameworks grounded in diffusion processes provide a unifying perspective that extends traditional graph neural network formulations while addressing limitations of standard message-passing mechanisms. Despite these advances, concerns remain regarding the fairness of such models, as they may propagate or amplify biases present in the data. In this work, we introduce a fairness-aware adaptation of graph-based diffusion by modifying the underlying Laplacian operator. Our approach incorporates multiple complementary transformations, including subspace projections, spectral adjustments, and frequency-based filtering, to mitigate bias-related components. Leveraging the intrinsic smoothing properties of graph diffusion, we provide a principled analysis of the resulting behavior and establish theoretical insights into fairness properties. We evaluate the proposed framework on both synthetic and real-world datasets, demonstrating that it achieves competitive performance while improving fairness metrics with limited additional computational cost.

Abstract

Explore similar work

Geometrical fairness in graph neural networks

FairGC: Fairness-aware Graph Condensation

Investigating the Effect of k-NN Preprocessing on Developing Graph Neural Networks: A Fairness-Based Perspective