Heart Disease is believed to be the number one killer globally, and its diagnosis has long been considered a very crucial problem. With the revolution of the modern world, it is very important to detect heart disease at its earlier stages so that patient treatment should be done effectively. Many previous researchers used Hybrid and Data Mining techniques to predict heart disease at its earlier stages, but they couldn't get the required results. The evaluation of the Machine learning and artificial intelligence research community mainly focused on these techniques to get better results. This research paper used six supervised machine learning classifiers like Random Forest, Support Vector Machine, Logistic Regression, Decision Tree, K-Nearest Neighbor and Naïve Bayes. We used two different datasets for the sample data in terms of attributes and values. We also used three different feature selection techniques to improve our accuracy by selecting the most important features. We first applied these machine learning classifiers to our proposed dataset without using the feature selection technique and computing the results. After that, we applied LASSOM, MRMR and MIFS techniques and derived the results with improved accuracies. In the end, we make a comparison table between the results that are computed with and without the feature selection technique. According to our experimental results we can say that the results accuracies computed with the feature selection technique are higher than those computed without feature selection techniques. Evaluation techniques like Confusion matrix, Accuracy, precision, Recall, F1 Score, PR Curve and ROC curve are used to measure the performance of our classifiers. So, we conclude that if we use any of these feature selection techniques, we can conclude better results and predict heart disease at its earlier stages with improved accuracy.