Key result
Ensemble machine learning models achieve the highest recall for clinical heart disease risk assessment.
Why the study?
Many heart attacks occur in people unaware of their risk despite presence of warning signs, necessitating improved risk assessment tools.
A proposed machine learning system combining clinical attributes and sensor data with Shapley value explanations aims to improve early heart disease risk detection and support clinical decision-making.
May support exploratory risk stratification; leaves open prospective validation before any clinical use.
Cardiovascular disease is the leading cause of death worldwide, and coronary heart disease accounts for a large share of these deaths. Many heart attacks occur in people who were unaware that they were at risk, even though warning signs such as high blood pressure, high cholesterol, abnormal ECG patterns, and reduced exercise tolerance were present beforehand. This paper presents a heart disease risk assessment system that uses machine learning to estimate a person's likelihood of heart disease from routinely measured clinical attributes and signals, and presents the result in a form that patients and doctors can understand. The system uses clinical attributes widely recorded in cardiology practice, including age, sex, type of chest pain, resting blood pressure, serum cholesterol, fasting blood sugar, resting electrocardiogram result, maximum heart rate achieved during exercise, exercise-induced angina, ST depression, slope of the peak exercise ST segment, number of major vessels seen on fluoroscopy, and thalassemia test result. From an electronics perspective, several of these inputs come directly from biomedical instruments, and the system can also accept heart rate readings from a pulse sensor and basic parameters from a single-lead ECG module for continuous monitoring. The data is cleaned and preprocessed by handling missing and inconsistent values, one-hot encoding categorical attributes, and standardising numerical measurements. Several classifiers are trained and compared, including logistic regression, k-nearest neighbours, a support vector machine, naive Bayes, a decision tree, random forest, gradient boosting, and extreme gradient boosting. Hyperparameters are tuned with stratified cross-validation, and the models are evaluated using accuracy, precision, recall, F1 score, and the area under the ROC curve, with recall given particular weight. The best model assigns each person a probability of heart disease and a low, medium, or high risk category. Shapley value explanations show which attributes raised or lowered the risk for that individual, for example a high ST depression or a low maximum heart rate. A web dashboard allows a clinician or health worker to enter patient data, view the risk and its explanation, compare repeated assessments over time, and read general guidance on lifestyle factors such as smoking, diet, and physical activity. The system is intended to support clinical decision making and preventive care, not to replace diagnosis by a cardiologist. It can help primary care doctors and health camps decide which patients need further cardiac investigation. Future work includes integration with wearable ECG devices for continuous risk tracking, validation on larger and more diverse patient populations, and the addition of signal-level features extracted from full ECG recordings.
No takes yet. Share an insight, caveat, or question.
1 B.Santhosh, 2 Divya, 3 Singireddy Pavani, 4 Singarapu Srija, 5 Rajurapu Naveen (2026) studied Heart disease. Machine learning-based risk assessment system vs. Standard risk scores and clinical judgement was evaluated on Model performance (accuracy, precision, recall, F1 score, ROC-AUC). Ensemble machine learning methods, such as random forest and gradient boosting, provided the strongest performance with high recall for heart disease risk assessment using clinical attributes.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: