What is the ROC curve? When to use it?
Classic MLLevel: IntermediateClassification
What is the ROC curve? When to use it?
Hints
- Write TP, FP, TN and FN for a small example.
- Separate ranking quality, threshold selection and probability calibration. Which error costs more?
Answer
A receiver operating characteristic (ROC) curve evaluates a binary scoring model across decision thresholds. Its vertical axis is TPR = TP/(TP + FN); its horizontal axis is FPR = FP/(FP + TN). Lowering the threshold usually increases both rates.
Use it to compare the sensitivity/false-alarm trade-off and inspect an operating region relevant to the application. It is not a probability-calibration curve and does not itself predict an outcome probability. ROC-AUC summarizes ranking over thresholds, but can hide poor precision when positives are rare. In that setting also inspect a precision-recall curve, and choose the deployed threshold using actual costs or constraints.
Source and licence: Alexey Grigorev & contributors — Data Science Interviews · CC-BY-4.0 · answer: Quantoros editorial