Model Architecture & Performance
Published:
Shpaner, L., Saitta, G., et al. BMC Urology (2026).
doi:10.1186/s12894-026-02072-x
We compared three supervised models: logistic regression, random forest, and support vector
machines to predict short-term complications after adult male circumcision. Data from 194
patients (≥18 years) at a Milan center (2023–2024), using age, BMI, blood loss, surgical
technique (traditional vs. laser), intraop vitals, and comorbidities.
Resampling & Training
- No resampling (SMOTE/ROS) was applied
- Stratified 10-fold cross-validation
- Standard preprocessing pipelines
Performance Metrics (SVM Best)
| Metric |
Point Estimate (95% CI) |
|---|
| AUC ROC | 0.907 (0.855–0.950) |
| Precision / PPV | 0.725 (0.623–0.826) |
| Average Precision | 0.832 (0.735–0.913) |
| Sensitivity / Recall | 0.862 (0.765–0.939) |
| Specificity | 0.860 (0.797–0.915) |
| F1-Score | 0.787 (0.706–0.857) |
| Brier Score | 0.105 (0.077–0.134) |
| Model Threshold (Point Estimate) | 0.238 |
Top Predictors (SHAP)
Intraoperative blood loss and surgical technique were the strongest predictors.
Notes & Next Steps
Strengths: rigorous CV, interpretability via SHAP.
Limitations: single-center data, no external validation.
Future: test on diverse populations and settings.
Hyperparameter Tuning
| Parameter |
Options |
Chosen |
|---|
| Kernel Type | linear, rbf, poly, sigmoid | rbf |
| Cost Parameter (C) | 0.0001 – 100 (log scale) | 100 |
| Gamma (γ) | 0.001, 0.01, 0.05, 0.1, 0.2, 0.5, scale, auto | auto |
| Resampling Method | None, SMOTE, ROS | None |
