The heart plays a vital role in living organisms. Heart-related disorders are more difficult to diagnose and predict, and errors in diagnosis and prognosis can have fatal consequences. A subfield of artificial intelligence, machine learning uses training from natural occurrences to predict any type of event with minimal support. In this work, we estimate the predictive power of machine learning algorithms that use several feature selection strategies to forecast heart disease. These algorithms include Select k best, Mutual Information Score, Recursive Feature Elimination (RFE), Support Vector Machine (SVM) estimator, and cross-validation. Random forest and gradient boosting were the predictive models used on the Framingham dataset for training and testing. This study uses the SMOTE technique for class imbalance problems. In our research, we got the highest accuracy for select k best feature selection modeled using random forest with an accuracy of 95.61% and gradient boosting with an accuracy of 95.43% considering hyperparameters.