Upload CSV data to generate ROC curves, compute AUC with confidence intervals, and find the optimal threshold by Youden index. Data never leaves your browser.
Try it out
Load example ROC/AUC data to see the full workflow
CSV must contain a binary label column (0/1) and a numeric score column. Drag & drop or click to upload.
When to use
Do not use for
An AUC of 0.5 means no better than chance on average, but check the ROC curve shape. A curve that crosses the diagonal can have AUC near 0.5 while being informative at certain thresholds. Always inspect the curve, not just the summary number.
The Youden index equally weights sensitivity and specificity. For screening tests, you may prefer a lower threshold (higher sensitivity). For confirmatory tests, choose a higher threshold (higher specificity). Use the threshold table to pick a clinically appropriate cutoff.
A point estimate of AUC = 0.85 could have a 95% CI of [0.72, 0.98] in a small sample. The CI width reveals whether your sample size supports meaningful conclusions about discrimination.
Your CSV must have a column with 0/1 labels (true class) and a column with continuous numeric scores. Rows with missing or non-numeric values are automatically excluded. The tool auto-detects common column names like y_true, label, y_score, and prob.
The ROC curve is constructed by computing sensitivity (TPR) and 1−specificity (FPR) at every unique score threshold. AUC is calculated via the trapezoidal rule. The standard error uses the Hanley-McNeil method, which accounts for correlated sensitivity/specificity values. The Youden-optimal threshold maximizes J = sensitivity + specificity − 1. All processing runs client-side; uploaded CSV data never leaves the browser.
Last validated 2026-04-05. Calculations are designed for planning and documentation support; verify procurement decisions against manufacturer specifications or institutional SOPs.
ConductScience ROC/AUC Calculator (v1.0). ConductScience, Inc. 2026. Available at: https://conductscience.com/tools/roc-auc-calculator
Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology. 1982;143(1):29-36. doi:10.1148/radiology.143.1.7063747
ROC analysis evaluates the discrimination ability of a binary classifier or diagnostic test with a continuous output. Key concepts:
• Threshold: The cutoff value above which a case is classified as positive • Sensitivity (TPR): Proportion of true positives correctly identified • Specificity (1 − FPR): Proportion of true negatives correctly identified • AUC: Summary measure of overall discrimination (0.5 = chance, 1.0 = perfect) • Youden index: Optimal threshold maximizing J = Sens + Spec − 1
The ROC curve is threshold-invariant — it shows performance across all possible thresholds, making it more informative than accuracy at a single cutoff.
The threshold table shows diagnostic accuracy metrics at each unique score value. Use it to:
• Find the Youden-optimal threshold (highlighted) — maximizes combined sensitivity and specificity • Choose a sensitivity-first threshold — for screening, find the lowest threshold that gives 90% sensitivity • Choose a specificity-first threshold — for confirmation, find the highest threshold that gives 95% specificity • Evaluate PPV/NPV — at your chosen threshold, considering the prevalence in your target population
Browse scientific equipment for your research
Browse ProtocolReference protocol for the next planning step.
Read guide ProtocolReference protocol for the next planning step.
Read guide ProtocolReference protocol for the next planning step.
Read guide