cost-sensitive learning

**Cost-Sensitive Learning** is a **machine learning framework that incorporates different misclassification costs for different classes or types of errors** — using a cost matrix to penalize certain errors more heavily, reflecting the real-world consequences of different types of misclassifications. **Cost-Sensitive Methods** - **Cost Matrix**: Define costs for each (true class, predicted class) pair — not all mistakes are equal. - **Weighted Loss**: Weight the loss function by class-specific costs: $L = sum_i c(y_i, hat{y}_i) cdot ell(y_i, hat{y}_i)$. - **Threshold Adjustment**: Modify the decision threshold based on the cost ratio. - **Meta-Learning**: Learn the cost weights from validation performance. **Why It Matters** - **Asymmetric Costs**: Missing a killer defect (false negative) is far more costly than a false alarm (false positive). - **Business Alignment**: Costs can reflect actual financial impact of each error type. - **Flexible**: Cost-sensitive learning is model-agnostic — applies to any classifier. **Cost-Sensitive Learning** is **pricing each mistake** — incorporating the real-world cost of different errors into the model's training objective.