This study aims to analyze systemic risk in the Iranian banking system during 2008–2023 by employing a hybrid approach that integrates Graph Neural Networks (GNN), financial shock propagation simulations based on the interbank dependency matrix (GVD), and Generalized Variance Decomposition (GVAR). The primary objective is to identify banks that act as systemic risk hubs and play a significant role in transmitting financial shocks. Real banking performance data were modeled as a directed network with a GVD threshold of 0.2. Using data from nine major banks and indicators such as return on assets, interbank investments, non-performing loans, and banking obligations, the study identified the network structure of banks and patterns of risk contagion. The results reveal that certain banks, including Eghtesad Novin, Tejarat, and Saderat, serve as central hubs with the greatest role in amplifying and spreading risk. Dynamic network analysis indicates relative stability in intra-systemic connections but heightened interdependencies following crisis periods, such as the 2011 sanctions. Findings further show that larger banks, due to higher network centrality, exert greater influence on system stability, while smaller banks are more vulnerable to common shocks. The proposed hybrid model demonstrated sufficient accuracy in identifying high-risk banks, making it an effective tool for financial policymakers. Overall, the results underscore the importance of targeted supervision of key banks and the design of precautionary policies to mitigate systemic risk contagion.