PREDICTIVE HEARTS: A MACHINE LEARNING FRAMEWORK FOR EARLY CARDIOVASCULAR DISEASE DIAGNOSIS

Abstract

Cardiovascular diseases (CVDs) are the leading cause of death worldwide, responsible for approximately 17.9 million fatalities each year. Early detection is crucial to reducing this global health burden. While machine learning (ML)-based diagnostic tools show considerable potential, many existing models face challenges with imbalanced datasets and limited accuracy in multi-class CVD staging. These models often underperform on minority classes due to data imbalance and insufficient model complexity. Additionally, the lack of intuitive, real-time user interfaces limits their practical adoption in clinical environments. This study introduces a machine learning-based system for early detection and staging of cardiovascular disease, featuring a Tkinter-based graphical user interface (GUI) that streamlines dataset handling, preprocessing, and model evaluation. The dataset used includes 2892 patient records with 13 features such as age, sex, chest pain type, and maximum heart rate. After applying label encoding and addressing class imbalance with SMOTE, the dataset expanded to 3710 records, ensuring 742 instances per class in the test set. Two classification models were employed: a Random Forest Classifier (30 estimators, maximum depth of 3) and an Ensemble Extra Trees Classifier (100 estimators, maximum depth of 100). The Random Forest model yielded suboptimal results, with a precision of 51.54%, recall of 52.95%, F1-score of 51.29%, and an accuracy of 53.50%, particularly struggling with Stages 1 and 2 (F1-scores of 0.41 and 0.34, respectively). In contrast, the Ensemble Extra Trees Classifier achieved excellent results, with precision, recall, F1-score, and overall accuracy all exceeding 97.9%, and per-class F1-scores ranging from 0.95 to 1.00. The system also includes a prediction module for classifying new patient inputs into CVD stages (0: No Disease, 1–4: Stages 1–4) and a comparative graph that visualizes model performance. The use of SMOTE effectively handled class imbalance, while the GUI enhanced usability, making the tool suitable for clinical deployment. Future developments may focus on incorporating more advanced models, real-time data processing capabilities, and explainable AI techniques to further increase the system’s accuracy, reliability, and trustworthiness in medical diagnostics.