An Improved Hybrid Approach for Proactive Healthcare Industry Risk Prediction: Integrating Graph Weighted Fusion, RF and RW-GCN.

Abstract

Risk prediction in healthcare industrial chains is critical for ensuring operational resilience and systemic stability. Traditional methods often focus on supply chain topology while overlooking enterprise-level indicators and the complex interdependencies among participants. To address these limitations, we propose IA-GWRG, a hybrid framework that integrates multi-relational graph modeling, key indicator selection, and subgraph-level learning for fine-grained enterprise risk prediction. Designed to align with emerging 6G-enabled intelligent infrastructures, IA-GWRG leverages ultra-low latency and ubiquitous sensing to enable real-time risk awareness. First, we introduce a salient graph feature extraction method that transforms a multi-relational industrial chain graph into a homogeneous structure using weighted fusion based on "competition-supply" meta-paths. A local-global-positional scheme is then applied to capture multi-scale structural features. Second, a key indicator identification module constructs a domain-specific 26-dimensional risk system and employs a random forest algorithm to select the most informative micro-level indicators, enhancing interpretability and reducing feature noise. Third, a node risk assessment method combines selected indicators and structural embeddings via subgraph-level random walk sampling and GCN-based learning, enabling precise and context-aware risk prediction. Extensive experiments on real-world healthcare industry chain datasets demonstrate that IA-GWRG consistently outperforms state-of-the-art baselines across multiple metrics. The results confirm its effectiveness and robustness in both predictive accuracy and 6G-oriented deployment scenarios.