Abstract
Heart diseases are considered one of the leading causes of death globally over the world. They are difficult to be predicted by a specialist physician as it is not an easy task which requires greater knowledge and expertise for prediction. With the variety of machine learning and deep learning algorithms, there exist many recent studies in the state of the art that have been done remarkable and practical works for predicting the presence of heart diseases. However, some of these works were affected by various drawbacks. Hence, this work aims to compare and analyze different classifiers, pre-processing, and dimensionality reduction techniques (feature selection and feature extraction) and study their effect on the prediction of heart diseases existence. Therefore, based on the resulting performance of several conducted experiments on the well-known Cleveland heart disease dataset, the findings of this study are: 1) the most significant subset of features to predict the existence of heart diseases are PES, EIA, CPT, MHR, THA, VCA, and OPK, 2) Naive Bayes classifier gave the best performance prediction, and 3) Chi-squared feature selection was the data mining technique that reduced the number of features while maintained the same improved performance for predicting the presence of heart disease.