@Article{cmc.2025.065423,
AUTHOR = {Saadia Tabassum, Fazal Muhammad, Muhammad Ayaz Khan, Muhammad Uzair Khan, Dawar Awan, Neelam Gohar, Shahid Khan, Amal Al-Rasheed},
TITLE = {A Machine Learning-Based Framework for Heart Disease Diagnosis Using a Comprehensive Patient Cohort},
JOURNAL = {Computers, Materials \& Continua},
VOLUME = {84},
YEAR = {2025},
NUMBER = {1},
PAGES = {1253--1278},
URL = {http://www.techscience.com/cmc/v84n1/61781},
ISSN = {1546-2226},
ABSTRACT = {Early and accurate detection of Heart Disease (HD) is critical for improving patient outcomes, as HD remains a leading cause of mortality worldwide. Timely and precise prediction can aid in preventive interventions, reducing fatal risks associated with misdiagnosis. Machine learning (ML) models have gained significant attention in healthcare for their ability to assist professionals in diagnosing diseases with high accuracy. This study utilizes 918 instances from publicly available UCI and Kaggle datasets to develop and compare the performance of various ML models, including Adaptive Boosting (AB), Naïve Bayes (NB), Extreme Gradient Boosting (XGB), Bagging, and Logistic Regression (LR). Before model training, data preprocessing techniques such as handling missing values, outlier detection using Isolation Forest, and feature scaling were applied to improve model performance. The evaluation was conducted using performance metrics, including accuracy, precision, recall, and F1-score. Among the tested models, XGB demonstrated the highest predictive performance, achieving an accuracy of 94.34% and an F1-score of 95.19%, surpassing other models and previous studies in HD prediction. LR closely followed with an accuracy of 93.08% and an F1-score of 93.99%, indicating competitive performance. In contrast, NB exhibited the lowest performance, with an accuracy of 88.05% and an F1-score of 89.02%, highlighting its limitations in handling complex patterns within the dataset. Although ML models show superior performance as compared to previous studies, some limitations exist, including the use of publicly available datasets, which may not fully capture real-world clinical variations, and the lack of feature selection techniques, which could impact model interpretability and robustness. Despite these limitations, the findings highlight the potential of ML-based frameworks for accurate and efficient HD detection, demonstrating their value as decision-support tools in clinical settings.},
DOI = {10.32604/cmc.2025.065423}
}