Stroke is one of the leading causes of death globally and is particularly prevalent in Indonesia. Early prediction of stroke is critical to reducing the risk of long-term disability and mortality. This study aims to build a stroke prediction model using the Support Vector Machine (SVM) classification method. The dataset used is sourced from Kaggle, containing 5,110 records with class imbalance. To address the imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) was applied during preprocessing. The study evaluates model performance across multiple data splits (70:30, 80:20, 90:10) and k-fold cross-validation values (k=5, 7, 10). The SVM was tested with various kernel types—linear, polynomial, and radial basis function (RBF)—along with parameter tuning for C, gamma, and degree. The results show that the polynomial kernel yielded the highest prediction accuracy of 92%. The model performance was evaluated using accuracy, precision, recall, and F1-score metrics.