Machine learning has recently drawn the attention of medical fields, with several cutting-edge machine learning frameworks that have been developed for heart disease prediction using diversified datasets. The ability of machine learning models to achieve more accuracy in predicting heart diseases and making them resilient to performance improvement is not yet achieved fully. This paper introduces a novel way to predict heart disease based on machine learning techniques and investigates the effectiveness of stacking classifiers. This topical research analyzes the data from popular datasets on heart disease to enhance prediction by using the combination of feature selection, management of outliers, and hyperparameters. In the current work, the performance of some key machine learning algorithms, such as Random Forest, XGBoost, and Gradient Boosting, is checked with the optimization of model performance increased using grid search and cross-validation, RF and Gradient Boosting models obtained the accuracy of 97%, respectively, also obtained accuracy of XGBoost reached 96% compared to other models. Our findings suggest that the stacking classifier illustrates superiority, and its accomplished accuracy is very high, close to nearly 98%, in comparison with traditional ML models. Our current work accentuates the potential of advanced ensemble methods in earlier detection and screening of heart disease and provides insights for more methodological advances and clinical applications, the proposed approach demonstrates robust predictive capability, paving the way for future research into more diverse and comprehensive datasets to validate and enhance the model’s efficacy.