GAT-RWOS: Graph Attention-Guided Random Walk Oversampling for Imbalanced Data Classification

TL;DR

GAT-RWOS is a novel graph-based oversampling technique that uses attention-guided random walks to generate synthetic minority samples, significantly improving classification in imbalanced datasets.

Contribution

It introduces a new oversampling method combining Graph Attention Networks with random walks to enhance minority class representation.

Findings

Outperforms existing oversampling methods on multiple datasets.

Effectively expands class boundaries while maintaining data distribution.

Improves overall classification accuracy on imbalanced data.

Abstract

Class imbalance poses a significant challenge in machine learning (ML), often leading to biased models favouring the majority class. In this paper, we propose GAT-RWOS, a novel graph-based oversampling method that combines the strengths of Graph Attention Networks (GATs) and random walk-based oversampling. GAT-RWOS leverages the attention mechanism of GATs to guide the random walk process, focusing on the most informative neighbourhoods for each minority node. By performing attention-guided random walks and interpolating features along the traversed paths, GAT-RWOS generates synthetic minority samples that expand class boundaries while preserving the original data distribution. Extensive experiments on a diverse set of imbalanced datasets demonstrate the effectiveness of GAT-RWOS in improving classification performance, outperforming state-of-the-art oversampling techniques. The proposed method has the potential to significantly improve the performance of ML models on imbalanced datasets and contribute to the development of more reliable classification systems.