Unbalanced Dataset

I am trying to make a node classification on a dataset of graphs I have. The problem is that the data is very imbalanced. Is it possible to use DGL to create a detector of anomalous nodes in a graph?

Thanks a lot!


How do you define anomalous nodes? For imbalanced data, you can considering focal loss or weighted cross entropy loss

The values of the node features are different, also they are not connected to other nodes similiarly to other nodes.