ASTHMA DIAGNOSIS USING OPTIMIZED ENSEMBLE STACKING AND SHAP-BASED FEATURE SELECTION: A HIGH-PRECISION MACHINE LEARNING PIPELINE

Asthma, a chronic respiratory condition affecting millions globally, demands early and accuratediagnosis to prevent severe health consequences. This study presents an advanced machine learning frameworkthat achieves exceptionally high prediction accuracy on an asthma dataset, utilizing a robust ensemble strategy,data resampling techniques, and SHAP-based interpretability. By combining LightGBM, CatBoost, andXGBoost in a Stacking Classifier with Logistic Regression as a meta-learner, and leveraging SMOTEENN forclass balancing, we achieved a cross-validation accuracy of 99.07%, test accuracy of 99.07%, and trainingaccuracy of 99.97%. The proposed methodology is benchmarked to exceed current state-of-the-art asthmaprediction models.