Glossary
Class Imbalance
Also called: imbalanced data
When one outcome is far more common than another in a dataset, which distorts both training and the interpretation of accuracy.
Most clinical conditions are rare relative to their screening population. A dataset that is 99 percent normal makes accuracy meaningless ... a model calling everything normal scores 99 percent ... and makes training difficult because the positive signal is sparse.
Where This Gets Misread
The usual mitigation is enriching the dataset with positive cases, which fixes training and breaks the reported performance figures. A model evaluated on a 50/50 enriched set has a positive predictive value that will not survive contact with real prevalence. Check what prevalence the reported metrics assume.