How do you handle imbalanced datasets in classification problems?

Imbalanced datasets require specialized techniques to prevent models from being biased toward majority classes.\n\n• Resampling: SMOTE for oversampling, random undersampling for majority class\n• Algorithm-level: Class weights, cost-sensitive learning, threshold tuning\n• Evaluation: Focus on precision, recall, F1-score, AUC instead of accuracy\n• Ensemble methods: Balanced bagging, boosting with class weights\n\nExample: Fraud detection with 1% positive cases. Apply SMOTE to generate synthetic fraud examples, use class_weight="balanced" in algorithms, evaluate with precision-recall curves, and optimize threshold based on business cost of false positives vs false negatives.