There are 3 levels of solutions to multiclass classification with imbalanced datasets: 1. data 2. algorithm 3. eval Weighted Cross entropy loss and Focal Loss are often used on algorithm level. Can work on these too, eventually, if needed