An Improved Framework for Cardiovascular Disease Prediction using Hybrid Ensemble Learning Soft-Voting Model

Cardiovascular diseases (CVDs) remain the leading cause of death globally, accounting for approximately 17.9 million deaths each year. While diagnostic technologies have advanced, traditional methods often fail to detect early-stage heart disease, particularly in asymptomatic individuals. This study presents an intelligent, interpretable machine learning framework designed to enhance early CVD prediction. guided by domain-specific medical knowledge. Two well-established datasets the Kaggle Cardiovascular Disease Dataset and the Framingham Heart Study Dataset were used for model training and evaluation. Machine learning models including Random Forest, Support Vector Machine, and neural networks were applied, along with SMOTE-Tomek Links to address class imbalance. A score-specific weighted soft voting mechanism was used to improve prediction across varying risk categories. The final model achieved high performance: Score 1 (low risk) reached 97.4% accuracy and 0.99% AUC; Score 2 (moderate risk) achieved 83.6% accuracy and 0.90% AUC; Score 3 (high risk) yielded 93% accuracy and 0.97% AUC. These results demonstrate the model's high precision, recall, and generalizability across diverse patient profiles. By combining ensemble learning, medical knowledge, and advanced risk stratification, the framework offers a clinically relevant tool for early detection and personalized intervention, supporting proactive cardiovascular healthcare.