F1-Score vs Accuracy for Imbalanced Datasets

Question: When would you prefer F1-Score over Accuracy as an evaluation metric? Key points: Accuracy is misleading on imbalanced datasets - a model predicting the majority class every time can score high accuracy while being useless. F1-Score (harmonic mean of precision and recall) better reflects performance on the minority/positive class, since it penalizes both false positives and false negatives. Prefer F1 (or precision/recall individually) whenever class distribution is skewed or the cost of false positives/negatives is asymmetric.