ML//Evaluation//class imbalance

When one class massively outnumbers the other, so a model can score well by mostly ignoring the rare class.

When one class massively outnumbers the other, so a model can score well by mostly ignoring the rare class.

Breaks accuracy and ROC-AUC; use PR-AUC and base-rate behavior instead.

Even a good model concentrates but does not purify: most alarms stay false because most cases are non-events.

Handled at training time with class-weighted cross-entropy, focal loss, or resampling.