A Multi-Class Heart Disease and Stroke Risk Classification Framework for Combined Assessment: A Machine Learning Approach

Abstract

This study introduces a combined multi-class machine learning system in predicting both heart disease and stroke risks. This was achieved by creating a harmonized dataset by merging two independent cardiovascular cohorts via conditionbased feature alignment and bias-controlled balancing to be able to make consistent risk modeling across diseases. The suggested framework, in contrast to the previous single-disease methodology, creates a single, explainable prediction pipeline that uses routinely available clinical biomarkers such as cholesterol, fasting glucose, and blood pressure to create detailed risk profiles of the individual patient. Several algorithms were compared and LightGBM demonstrated better performance with stroke and heart disease recall rate being 0.91 and 0.90 respectively with a total of 89% accuracy and an AUC of approximately 0.99. In addition to the predictive performance, interpretability studies based on SHAP values demonstrated similar clinical relevance of the major predictors and also showed cross-disease feature interactions. These results provide an indication that an integrated, interpretable, and data-efficient learning system has the potential to support screening and risk of cardiovascular disease in the early stages of care and inform preventive care in a real healthcare environment.