Improving Anomaly detection in Heterogen eous Graphs using GAN-Enhanced Explainability

Amid the era of rapid integration of Graph Neural Networks (GNNs) into diverse applications, such as anomaly detection in heterogeneous graphs, the pursuit of interpretable and dependable decision-making models has gained paramount significance. The comprehension of the rationale behind GNN predictions proves vital for instilling trust in their outcomes. While existing GNN explanation methods have made progress in this domain, their ability to provide precise and authentic explanations remains limited. To address these deficiencies, a novel approach named “GANEnhanced Explainability” (GEE) is introduced, built upon the foundation of Generative Adversarial Networks (GANs). GEE comprises a dual-component architecture: a generator responsible for producing explanations and a discriminator facilitating the refinement of the explanation generation process. Additionally, this research extends this foundation by incorporating innovative visualization techniques to augment the interpretability of GNN-based anomaly detection, thereby forging a new path in the realm of explainable AI for intricate graph structures.