EXPLAINABLE GRAPH NEURAL NETWORKS IN EPIDEMIOLOGICAL SYSTEMS: CURRENT ADVANCES AND RESEARCH FRONTIERS IN SENSOR-DRIVEN ZOONOTIC DISEASE MODELLING

Recent advances in explainable graph neural networks (XGNNs) have opened new possibilities for modelling complex, dynamic relationships in epidemiological systems. This study synthesizes theoretical, empirical, and technical developments at the intersection of graph representation learning, explainable artificial intelligence (XAI), and computational epidemiology, focusing on sensor-driven zoonotic disease modelling. Through a qualitative comparative review of current literature, the study identifies four convergent research dimensions: the algorithmic potential of GNN-based frameworks to represent disease transmission dynamics; hardware-aware and scalable model design that supports real-time inference; sensor data integration strategies that improve contextual accuracy and data governance; and emerging interpretability techniques that enhance clinical and public health trust in AI-assisted surveillance systems. The synthesis highlights that while GNNs effectively capture relational dependencies across heterogeneous epidemiological networks, their deployment remains constrained by computational scalability, data quality, and limited interpretability. Explainability methods such as attention mechanisms, GNNExplainer, and GraphLIME are increasingly bridging this gap by revealing the subgraph structures and features most responsible for model predictions. The study proposes a hardware-aware, explainable GNN pipeline that integrates multi-sensor data, interpretable inference, and privacy-preserving mechanisms to support early zoonotic outbreak detection and transparent decision-making. This framework positions XGNNs as a foundation for next-generation epidemiological intelligence systems capable of uniting algorithmic precision with human interpretability.