Unlock: Class Imbalance and Resampling

When class frequencies differ dramatically, standard accuracy is misleading. Resampling, cost-sensitive learning, and threshold tuning restore meaningful evaluation and training.