Support Vector Machine (SVM) finds the best-separating hyperplane to classify data points into two categories which are quantitative and qualitative features. This research aims to predict heart failure using SVM, focusing on classification accuracy and algorithm efficiency. This research addresses a review of previous studies that identified gaps. The research methodology involves a framework with four key components: Data Preparation, Data Pre-Processing, Feature Selection, and Classification; using a dataset specifically on heart failure. This research proposes using Recursive Feature Elimination (RFE), Random Forest Importance (RFI), and LASSO to improve attribute reduction and classification speed, aiding timely patient treatment. This study focuses solely on SVM and feature selection methods to determine accuracy, precision, and recall in predicting heart failure. Linear kernel and Radial Basis Function (RBF) kernel hyperparameters are implemented and the results from the validations indicated that the SVM + RFE and SVM + LASSO models consistently outperformed others, achieving the highest average accuracy, precision, and recall scores in both linear and RBF kernels. This comprehensive approach confirmed the superiority of feature selection combined with hyperparameter optimization in enhancing SVM model performance. The highest accuracy obtained is 86.16% at the RBF kernel of both SVM + RFE and SVM + LASSO.