Area Under the Curve (AUC) summarizes the area beneath a receiver operating characteristic (ROC) curve. It measures how well a classifier separates positive from negative examples across decision thresholds: larger values indicate better ranking performance. The ROC curve plots true-positive rate against false-positive rate, while the diagonal represents chance-level behavior. section 4.3 section 5.2.3
Use AUC as a practical comparison metric when evaluating classifiers or link-prediction models:
- Compare models on the same test split using predicted probabilities rather than only hard class labels. section 4.3 section 4.3.1
- Inspect the ROC curve alongside its AUC; a curve near the diagonal signals weak separation, even when another metric appears acceptable. section 4.3.1
- Treat AUC as a ranking-oriented summary across thresholds, not as a single operating-point measure. section 4.3
- For graph autoencoders, calculate AUC on validation and test edge predictions after training, with nonexistent edges represented as negative examples. section 5.2.3
Baseline comparison
On the same tabular node-feature split, logistic regression achieves about 75.90% ROC AUC, an MLP reaches 85.93%, and XGBoost reaches 94.17%. The ROC plots make the ranking difference visible: XGBoost lies above the other baseline curves, while logistic regression remains relatively close to chance. section 4.3.1 section 4.3
from sklearn.metrics import roc_curve
fpr, tpr, _ = roc_curve(ytest,ypred) #1
plt.figure(1)
plt.plot([0, 1], [0, 1])
plt.plot(fpr, tpr)
plt.xlabel('False positive rate')
plt.ylabel('True positive rate')
plt.show()
Plot false-positive rate against true-positive rate to inspect the ROC curve and its relationship to the chance diagonal. section 4.3.1
