Prediction of Myocardial Infarction Risks Using Support Vector Machine and Hyperparameter Optimization

Abstract

Cardiovascular diseases, notably myocardial infarction (MI) remain as the primary cause of global mortality, with millions of deaths each year, often attributed to delayed diagnosis and intervention. This study focused on predictive modelling for MI risks using Support Vector Machine (SVM) algorithm, with its hyperparameter tuned by both GridSearchCV and RandomizedSearchCV. Data extracted from UCI Heart Disease Dataset underwent comprehensive preprocessing pipeline including normalization and Chi-Square feature selection to identify key predictors. The predictive accuracy, precision, recall, F1-score, and Area Under the Receiver Operating Characteristic Curve (AUROC) were evaluated. Both the Grid Search and Randomized Search tuned models achieved a superior accuracy of 84.21%, with the Grid Search model yielding a slightly higher AUROC of 0.9137, confirming the effectiveness of hyperparameter tuning. This outcome highlights the superior discrimination and classification performance in predicting MI. This finding demonstrates the potential of machine learning to enable timely, accurate diagnosis of cardiovascular diseases and therefore improve patient care.