Beyond traditional models: Jaya-optimized ensembles for accurate heart disease prediction

Heart Disease (HD) stands as the foremost reason for mortality all over the world for both men and women. Millions of people are affected worldwide every year, resulting in numerous fatalities. Timely and precise detection is essential for enhancing patient survival rates and potentially preventing further complications. To avoid these situations, we herein present novel and practical ensemble methods that include different machine learning (ML) and Jaya optimization techniques for HD classification. The key benefits of the Jaya optimization method include its simplicity in implementation, faster convergence, and the absence of algorithm-specific parameter requirements.

This research method includes six major steps: (i) two different HD dataset descriptions, (ii) preprocessing of individual data, (iii) initial prediction with traditional ML models, (iv) final prediction using proposed Jaya-optimized Stack Ensemble (J-oSE) method, and (v) comparative performance evaluations using confusion matrix (cm), receiver operating characteristic (roc) curves and 10-fold cross-validation (cv) with a 95 % confidence interval, and (vi) statistical paired t-test to evaluate the significance of proposed method. The novelty found for the proposed method lies in the successful classifications of features that are more relevant to HD.

The proposed method at 10-fold CV with a 95 % confidence model achieved a significant accuracy of 93.55 % for the first dataset and 84.88 % for the second dataset, surpassing the outcome of other predefined methods like Voting Ensemble, XGBoost, and the baseline stacking model. Additionally, the found p-value as '1.0' for a proposed J-oSE method with Jaya optimization and '0.146' without Jaya optimization signifies that the data is perfectly distributed as normal for the first case and approximately normal for the second case, as p>0.05. The result for the T-statistic is 46.89; a high + ve t-value signifies that our proposed J-oSE method with Jaya optimization outperforms the ensemble method without Jaya optimization.

However, incorrect positive diagnoses of HD can result in unnecessary anxiety and therapy; wrong negative diagnoses can potentially be fatal. These goals are balanced by the Jaya + stacking optimization method to guarantee the greatest results throughout every key metric. Thus, we can say that our proposed method is more reliable and robust, potentially expanding the boundaries of predictive performance.