I’m trying to do node classification using R-GCN. I have two classes that my nodes beling to with extreme class imbalance (minority class making up only 8% of data points). How do people deal with such class imbalance in node classification?
Have you tried some common techniques for imbalance tabular data classification?
If the problem permits introducing data of rare class then techniques like SMOTE or AdaSyn are good option. And if not, then something like class weighting needs to be implemented in dgl’s backward propogation. I am not sure if it is already implemented or not.
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.