Oversampling, then applying a train-test split? Jesus, that's like machine learning 101. But then again, I see a lot of questionable practices in the application of ML in biology.