A Machine Learning Framework for Heart Disease Prediction With Feature Selection and Hyperparameter Optimization Across Benchmark and Real-World Datasets

Heart disease is a major cause of death worldwide, and thus there is a need to have good, sound, and clinically usable predictive models. In this paper, a machine learning model is developed for the prediction of heart diseases using both a benchmark dataset (UCI-Statlog) and a real-world dataset obtained from the Ayush Multi-Specialty Hospital and Research Centre (AMSHRC), Vijayapura, Karnataka, India. Five machine learning methods - Decision Tree, Random Forest, Logistic Regression, XGBoost, and AdaBoost - were tested under three different experimental conditions: without feature selection, Chi-Square feature selection, and Analysis of Variance F-Test feature selection. Hyperparameters were optimized using Grid Search Cross-Validation. The Random Forest model achieved 88% accuracy without feature selection and 86% accuracy with Chi-Square selection, and 85% with Analysis of Variance F-test feature selection methods applied to the UCI-Statlog dataset, which means that feature selection did not improve the performance of the model on this dataset. Across all experimental configurations, XGBoost achieved the highest accuracy of 96% and an area under the curve of 0.99 on the AMSHRC dataset, while Logistic Regression exhibited fairly consistent predictive performance despite increasing clinical complexity in the dataset, highlighting the robustness of the evaluated models. The proposed framework provides a realistic and generalizable decision-support model to determine the risk of heart diseases in the early stages. In conclusion, this study is based on the prediction of heart diseases using machine learning, feature selection, and real-world data along with grid search cross-validation and classification.