سامانه همایش های دانشگاه آزاد اسلامی واحد نجف آباد

صفحه اصلی / اولین همایش بین المللی تحول خدمات سلامت: تلفیق مراقبت، درمان، بازتوانی و امیدبخشی

A Comparative Study of Machine Learning Algorithms for Cardiovascular Disease Prediction: Toward Intelligent Clinical Decision Support

نویسندگان :

Mohammad Rouhi¹ Sajjad Ghatei² Zohreh Arabi³

1- دانشگاه پیام نور 2- دانشگاه ملی مهارت 3- دانشگاه ملی مهارت

کلمات کلیدی :

بیماریهای قلبی عروقی،یادگیری ماشین،XGBoost،پشتیبانی از تصمیم گیری بالینی،تبیین پذیری SHAP،پرونده های الکترونیک سلامت

چکیده :

Background: Cardiovascular diseases are among the leading preventable causes of mortality worldwide, accounting for approximately 17.9 million deaths annually [1]. Despite the use of tools such as the Framingham Risk Score, their reliance on linear regression models limits their ability to identify complex and non-linear relationships among clinical variables [2]. Methods: Six machine learning algorithms, including Logistic Regression, Decision Tree, Random Forest, Support Vector Machine, XGBoost, and Deep Neural Network, were trained and evaluated on data from 14,820 adults from the Cleveland Heart Disease Database and MIMIC-III [3]. SMOTE was applied exclusively to the training portion of each cross-validation iteration; the pre-SMOTE class ratio was 37.9% positive and 62.1% negative. RFECV reduced the number of features from 47 to 28, and stratified 10-fold cross-validation was performed [4]. Findings: XGBoost achieved the best performance, with an AUC of 0.97, accuracy of 96.4%, and F1-score of 0.96. Compared with the Framingham Risk Score (AUC=0.78 and accuracy=74.3%), it reduced the false-negative rate from 16.2% to 4.8%, representing a 70% relative improvement. The SHAP values for systolic blood pressure, LDL cholesterol, age, and left ventricular ejection fraction were 0.312, 0.287, 0.265, and 0.241, respectively [5]. Conclusion: XGBoost combined with SHAP can improve the predictive performance and interpretability of cardiovascular disease risk assessment. The study limitations include the single-centre origin of the MIMIC-III data, sex imbalance, and the absence of independent external validation.