Metric

Definition

Interpretation

Accuracy

(TP + TN)/(TP + TN + FP + FN)

Overall correct predictions may be misleading with imbalanced classes

AUC (Area under ROC Curve)

The probability that a randomly chosen obese child is ranked higher than a non-obese child

0.5 = random; 0.7 - 0.8 = acceptable; >0.8 = good discrimination

Precision

TP/(TP + FP)

Proportion of predicted positive cases that are truly obese

Recall (Sensitivity)

TP/(TP + FN)

Proportion of actual obese cases correctly identified

F1-Score

2 × (Precision × Recall)/(Precision + Recall)

Harmonic mean of precision and recall

Calibration

Agreement between predicted probabilities and observed frequencies (e.g., Hosmer-Lemeshow test, calibration plots)

Well-calibrated models have predicted risks matching actual event rates