Short Definition
A ROC curve visualizes the trade-off between true positive rate and false positive rate across decision thresholds.
Definition
The Receiver Operating Characteristic (ROC) curve plots a classifier’s performance by showing how the true positive rate (TPR) changes with the false positive rate (FPR) as the decision threshold varies. Rather than evaluating a model at a single threshold, the ROC curve summarizes behavior across all possible thresholds.
The ROC curve is commonly used to compare classifiers independently of a specific operating point.
Why It Matters
Many models output probabilities, not hard class labels. The ROC curve shows how sensitive performance is to threshold choice and helps determine whether a model can separate classes at all.
It is especially useful when class distributions change or when the optimal threshold is not known in advance.
How It Works (Conceptually)
- The model outputs scores or probabilities
- A threshold is swept from high to low
- For each threshold, TPR and FPR are computed
- Points are plotted to form a curve
A curve closer to the top-left corner indicates better class separability.
Key Rates
True Positive Rate (TPR) = TP / (TP + FN)
False Positive Rate (FPR) = FP / (FP + TN)
Minimal Python Example
# conceptual exampletpr, fpr = compute_rates(y_true, y_scores, threshold)
Common Pitfalls
- Interpreting ROC curves without considering class imbalance
- Assuming a good ROC curve implies good calibration
- Ignoring the practical decision threshold
- Comparing ROC curves across fundamentally different datasets
Related Concepts
- AUC
- Precision
- Recall
- Confusion Matrix
- Evaluation Metrics
- Decision Thresholding
Neural Network Lexicon 