Machine Learning
Imbalanced Data
When one class is rare, accuracy lies.
What to do
- Use precision/recall/F1 or PR-AUC.
- Try
class_weight="balanced"for many sklearn models. - Choose threshold based on cost (not just 0.5).
from sklearn.linear_model import LogisticRegression model = LogisticRegression(max_iter=300, class_weight="balanced")