Heart Disease Detection Using Machine Learning

Abstract

cardiovascular diseases (CVDs) persist as the main cause of death in the world, and this has been the requirement of the creation of sound automated diagnostic systems to enable early intervention. The non-linear nature of clinical datasets and their high-dimensional noise is also a challenge to the existing traditional diagnostic methodologies. The effectiveness of traditional machine learning standards, which are Support Vector Machines (SVM) and Random Forest (RF), which are the current systems under consideration in this study, is evaluated in this research. Although these models are fundamental, they tend to level off in terms of performance in case of sparse or skewed electronic health records. In order to overcome such shortcomings, we propose a high-performance framework utilizing Extreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LightGBM). The proposed system leverages an iterative gradient-based learning approach and leaf-wise tree growth to minimize error residuals and enhance computational efficiency. Experimental results indicate that the proposed boosting architecture significantly outperforms existing SVM and RF benchmarks, achieving a superior predictive accuracy of approximately 95.4%. This study demonstrates that transitioning from bagging and margin-based models to advanced gradient boosting frameworks can substantially improve the reliability of automated cardiac risk assessment.