Improving Heart Disease Risk Prediction using Stacking Classifier

In the medical field, early cardiac disease prediction is extremely difficult, yet precise models can greatly enhance patient outcomes. An enhanced stacking ensemble classifier for the prediction of heart disease is proposed in this paper. Three Level 0 base learners XGBoost, Random Forest, and LightGBM are integrated into the model to identify various patterns and connections in clinical data. A Level 1 meta-classifier then combines and processes their probabilistic outputs, learning the best weighting for each base model to get the final prediction. The dataset used in this study include important clinical characteristics like cholesterol, blood pressure, and age. The Synthetic Minority Over-sampling Technique (SMOTE) was used to combat class imbalance, while RandomizedSearchCV was used to optimize hyperparameters and improve model performance. The suggested ensemble outperforms individual classifiers, according to experimental results, with an accuracy of 0.9783 and a ROC-AUC score of 0.9897. The proposed method provides a better solution in early detection of heart disease and overall healthcare outcomes.