This project addresses class imbalance in graph classification. We propose a sampling-based graph-of-graphs framework that rebalances the learning process by constructing informative subgraph samples, enabling more robust classification on skewed graph datasets.