M.S. Applied Data Science - Capstone Chronicles 2025

16

Table 2 Optimal Hyperparameter Configurations for Each Model in group A. Model Optimal Hyperparameters Logistic Regression C=10, penalty=l2, solver=lbfgs Random Forest max_depth=20, min_samples_split=2, n_estimators=100 XGBoost learning_rate=0.2, max_depth=7, n_estimators=200 SVM C=10, kernel=rbf MLP activation=tanh, alpha=0.0001, hidden_layer_sizes=(100, 50) Table 3 Optimal Hyperparameter Configurations for Each Model in group B. Model Optimal Hyperparameters Logistic Regression C=0.1, penalty=l2, solver=lbfgs Random Forest max_depth=20, min_samples_split=2, n_estimators=200 XGBoost learning_rate=0.2, max_depth=7, n_estimators=200 SVM C=10, kernel=rbf MLP activation=tanh, alpha=0.0001, hidden_layer_sizes=(100, 50)

of both overall predictive accuracy and the ability to correctly classify cases of metabolic syndrome. In Group A, XGBoost achieved the highest overall performance, with accuracy, precision, recall, and F1-score all at 0.90 and an ROC-AUC of 0.96. Random Forest followed closely with values ranging 86-87 but still a strong ROC-AUC of 0.95. The MLP neural

5.2 Model Performance Model performance was assessed on the test set using accuracy, precision, recall, F1-score, and ROC-AUC (Tables 4 and 5). Accuracy and ROC-AUC are reported directly, while precision, recall, and F1-scores represent weighted averages across both classes to account for class imbalance. These complementary metrics allowed for evaluation

167