Hybrid Machine Learning Based Heart Disease Detection

Accurate and non-deceptive forecast of cardiovascular disease (CVD) may be the most important factor in enhancing clinical outcomes and keeping timely interventions on track. In recent years, ensembling have been highly effective in predictive modeling have been developed by combining the various different base learners to create more accurate and generalized models. Among these, stacking -a technique that integrates the prediction of various classifiers using a final classifier (meta learner) has shown significant promise. This study delves into a compare- tive analysis of various predictive algorithms such as Logistic Regressor, K Nearest Neighbor Classifier, Decision Trees, and Ran domized Forest techniques are utilized in data science against a sophisticated stacking ensemble framework. The proposed model employs a combination of RF, DT, and KNN as base learners, while a Ridge Classifier acts as the meta-learner to aggregate and fine-tune the predictions. Through a rigorous evaluation on a publicly available heart.csv dataset, the Random Forest model has a notable accuracy of 88.52%, standing out as the best-performing single algorithm. However, the stacking ensemble outperformed all individual models, achieving an improved accuracy of 90.16%. The outstanding execution of stacking ensemble can be referred to its ability to reduce overfitting, mitigate bias, and capture complex data patterns. These results emphasize the potential of using diverse base models and a strategically chosen meta-learner to enhance predictive performance. The study provides valuable insights into designing optimized ensemble models for clinical risk assessment, ultimately contributing to better decision support and patient management in CVD diagnosis.