<p>Mpox, a zoonotic disease of growing global concern, presents significant public health challenges because of the complexity of its transmission dynamics. The need for innovative tools to model and control outbreaks has become increasingly evident in the context of recent global health crises. This study investigates the dynamics of mpox outbreaks through the application of Graph Convolutional Networks (k-GCN model), focusing on two epidemiologically distinct African contexts: the Republic of the Congo (endemic Clade I, zoonotic transmission) and Nigeria (re-emerging Clade II, human-to-human transmission). We conducted simulations using a compartmental model that integrates graph-based contact networks and epidemiological parameters specific to each country. Five intervention scenarios, ranging from no intervention to combined vaccination, quarantine, and rodent culling, were evaluated. A k-GCN model was then trained to predict node classes for the following year based on simulation data, and explainability techniques (saliency maps and integrated gradients) were used to identify the most influential transmission pathways. The results indicate that the k-GCN model captures relevant temporal and structural dependencies in the simulated graph-organized data. Among the simulated scenarios, combined strategies including targeted vaccination, quarantine, and reservoir-directed rodent control produced the largest reductions in human infections and epidemic burden. Explainability analysis suggested that human-rodent interactions were influential during the early epidemic phase, with a gradual shift toward human-to-human transmission over time. The study supports the potential usefulness of combining contact-based models with explainable graph neural networks for scenario-based epidemic analysis. The proposed approach may help compare intervention strategies and generate evidence for public health decision-making, provided that results are interpreted within the assumptions and calibration limits of the model.</p>