ROC-AUC Curve Analysis

ROC-AUC analysis evaluates how well a score or model distinguishes between two classes by plotting sensitivity against 1 − specificity across all thresholds and summarizing the curve by its area.

ROC-AUC 분석은 모든 임계값에서 민감도와 1−특이도를 그려 점수나 모형이 두 범주를 얼마나 잘 구분하는지 평가하고 곡선 아래 면적으로 요약하는 방법이다.

How do I evaluate how well a test or model separates positive and negative cases?

검사나 모형이 양성과 음성 사례를 얼마나 잘 구분하는지 어떻게 평가하나요?

ROC curve analysis plots the true positive rate against the false positive rate at every classification threshold, and the area under the curve (AUC) summarizes discrimination across thresholds. It fits evaluating diagnostic tests, risk scores and binary classifiers, where an AUC of 0.5 indicates chance-level discrimination.

When to use ROC-AUC Curve Analysis

• Evaluating a diagnostic test or risk score against a binary reference standard
• Comparing the discrimination of two or more binary classifiers
• Choosing a cutoff that balances sensitivity and specificity

When to use something else

• You need to know whether predicted probabilities match observed rates: use a calibration plot
• Classes are highly imbalanced and positives matter most: also report a precision–recall curve

Assumptions and requirements

• Binary reference outcome
• A continuous or ordinal score or predicted probability
• Independent observations
• Evaluation on held-out or cross-validated data when the model was fitted on the same sample

What the analysis reports

• ROC curve
• AUC with confidence interval
• Optimal threshold (e.g., Youden index)
• Sensitivity and specificity at chosen cutoffs
• DeLong test for comparing AUCs

How to report it

The model discriminated well between [positive] and [negative] cases, AUC = [value], 95% CI [[lower], [upper]].

Related methods

• Binary Logistic Regression
• Precision-Recall Curve
• Calibration Analysis

Run ROC-AUC Curve Analysis on Quorant