The Early prediction of heart disease at its primary stage, based on symptoms, poses a significant challenge for healthcare organizations worldwide. Machine learning algorithms play a crucial role in predicting and diagnosing heart disease. This paper empirically investigates the application of various types of single machine learning algorithms—Bayes Network, Neural Network, Decision Tree, JRIP—along with one hybrid algorithm (Vote) for heart disease prediction. The study utilizes the heart failure prediction dataset containing 11 features and 918 records. Additionally, two types of feature selection methods— subset and ranking—are employed, each including two techniques: best first, greedy stepwise, Chi-squared, and information gain, respectively. Our empirical findings reveal that the predictive performance of the Vote algorithm surpasses that of other single algorithms. Furthermore, the Vote algorithm outperforms previous studies in most evaluation measures. Concerning feature selection, all methods consistently choose eight attributes, indicating the exclusion of only three attributes—RestingBP, FastingBS, and RestingECG. However, the removal of these attributes does not significantly impact the performance of the prediction algorithms.