Table 1.
Patients characteristics.
Table 2.
Phase 1 main performance of models in classification of positive vs. negative appendectomy, with corresponding confidence interval. Abbreviations: SVM = Support Vector Machine; KNN = K-Nearest Neighbors; MLP = Multi-Layer Perceptron; ROC AUC = Receiver Operating Characteristic Area Under the Curve; CI = Confidence Interval.
Table 3.
Phase 1 additional performance metrics. Abbreviations: SVM = Support Vector Machine; KNN = K-Nearest Neighbors; MLP = Multi-Layer Perceptron; PR AUC = Precision–Recall Area Under the Curve; MCC = Matthews Correlation Coefficient; G-Mean = Geometric Mean; SVM = Support Vector Machine; KNN = K-Nearest Neighbors; MLP = Multi-Layer Perceptron.
Fig 1.
SVC Receiver Operating Characteristic (ROC) Curves for phase 1: With and Without 5-Fold Cross-Validation.
Fig 2.
Comparison of Model Performance Metrics: F1 Score, Accuracy, AUC, Precision, and Recall.
Fig 3.
SHAP Value Plots: Individual Impact and Mean Impact on Model Output.
Fig 4.
SVC Receiver Operating Characteristic (ROC) Curves for phase 2.
Fig 5.
Calibration Curves for Phase I Appendicitis Detection Model.
Calibration plot comparing uncalibrated and isotonic-calibrated probability estimates. The x-axis represents predicted probabilities, and the y-axis reflects observed outcome frequencies. The calibrated model demonstrates improved alignment with the diagonal reference line, indicating enhanced probability reliability and reduced Expected Calibration Error (ECE).
Fig 6.
Operating Threshold Performance Across Probability Cutoffs.
Threshold analysis showing sensitivity, specificity, precision, F1-score, and accuracy across probability thresholds ranging from 0.10 to 0.90 (step = 0.05). The optimal clinical operating point (threshold = 0.35) satisfies both predefined criteria: maximum F1-score and minimum sensitivity ≥90%, achieving a balanced trade-off between case detection and false-positive rate.