A Multi-Stage Machine Learning Technique for Heart Disease Prediction with Cluster-Aware Classification

Heart disease is one of the biggest public health risks in the world, and current predictive models are typically hampered by feature reliance and inconsistent data. The present study introduces a novel multi-stage machine learning technique incorporating cluster-sensitive classification to increase predictive accuracy. The method starts with anomaly detection and data augmentation to handle data-quality problems. Feature selection was used to preserve the most informative features, followed by feature augmentation through Gaussian Mixture Model (GMM) clustering, which was identified as the best clustering method in augmenting the feature set. SMOTE was utilized to address class imbalance, and stratified cross-validation was used to maintain model generalizability. Various classifiers were compared, and LightGBM attained the highest accuracy of 86.77%. The findings highlight the capability of combining unsupervised learning algorithms with supervised classification techniques to enhance the data-driven and robust characteristics of heart disease diagnosis models. This method opens the door for more consistent data-driven approaches to healthcare prediction problems.