XGBoost and SHAP-Based Analysis of Risk Factors for Hypertension Classification in Korean Postmenopausal Women

Figures

• 

  Figure 1. Methodology for the analysis of the hypertension classification using the 2022–2023 Korean National Health Insurance Service health examination dataset.

• 

  Figure 2. Heatmap of Pearson correlation coefficient for measured variables.

• 

  Figure 3. Confusion matrix for hypertension classification performance.

• 

  Figure 4. Comparison of hypertension prediction performance.

• 

  Figure 5. ROC curve for hypertension classification performance.

• 

  Figure 6. Log loss and accuracy changes during XGBoost training for hypertension classification.

• 

  Figure 7. (a) SHAP feature importance based on mean absolute SHAP values: age and WC were the top-ranked features. (b) SHAP summary plot for hypertension classification based on XGBoost, showing the direction and magnitude of each feature’s impact. Red means higher and blue means lower impact on hypertension risk.

• 

  Figure 8. SHAP force plots for a 55-year-old woman showing a predicted SBP (a) of 139.36 mmHg and DBP (b) of 81.55 mmHg.

• 

  Figure 9. SHAP force plots for a 65-year-old woman showing a predicted SBP (a) of 136.60 mmHg and DBP (b) of 81.78 mmHg.

• 

  Figure 10. SHAP force plots for a 75-year-old woman showing a predicted SBP (a) of 149.56 mmHg and DBP (b) of 87.51 mmHg, exceeding hypertension thresholds.

• 

  Figure 11. SHAP force plots for a case with WC = 64 cm, showing a predicted SBP (a) of 105.33 mmHg and DBP (b) of 70.75 mmHg.

• 

  Figure 12. SHAP force plots for a case with WC = 74 cm, showing a predicted SBP (a) of 140.42 mmHg and DBP (b) of 77.17 mmHg. At this threshold, SBP exceeds the hypertension diagnostic criteria. This scenario suggests that WC near 74 cm marks a critical inflection point for hypertension risk.