The Effect of SMOTE and Optuna Hyperparameter Optimization on TabNet Performance for Heart Disease Classification

Heart disease is a medical condition affecting the cardiovascular system, disrupting blood circulation and reducing cardiac function efficiency, which can lead to severe health complications. Early diagnosis of heart disease has become increasingly crucial as delayed detection can significantly impact patient outcomes and survival rates. While numerous studies have explored various approaches for heart disease classification, challenges related to data imbalance and improper parameter settings remain persistent issues that affect model performance. This research evaluated the effectiveness of combining TabNet with SMOTE and optuna hyperparameter optimization for heart disease classification. We conducted four experimental scenarios using a heart disease dataset with 303 instances: baseline TabNet, baseline TabNet with SMOTE, TabNet with Optuna, and TabNet with both SMOTE and Optuna. Results demonstrated that applying SMOTE alone to TabNet decreased model performance (accuracy from 85.24% to 77.04%, AUC from 0.89 to 0.83). However, when combining SMOTE with Optuna hyperparameter optimization, we achieved optimal performance with 90.16% accuracy, 93.33% precision, 87.50% recall, 90.32% F1-score, and 0.93 AUC. This represented a significant improvement over other configurations and several previous classification approaches. The integration of SMOTE with Optuna optimization provided an effective framework for heart disease classification that outperformed traditional methods particularly in discriminative capability as evidenced by the superior AUC score.