You are training a binary classifier where the positive class is rare, and a standard model reports high accuracy while missing many of the minority cases. You need a principled way to train and evaluate the model.
How would you handle a dataset that is heavily imbalanced?
