Primary Prediction of Heart Disease using Machine Learning Algorithms and SMOTE

The cardiovascular disease is one of the largest field where Artificial Intelligence and Machine Learning tools can be used to help in early prediction and diagnosis. As the first concern of doctors is to make the one percent precise diagnosis, these tools can be more than helpful. Lately, Physicians and searchers are performing systems capable of predicting the occurrence of such diseases. In this paper, we aimed to build a heart disease prediction system by conducting a comparison over two Machine Learning Classifier: Logistic regression (LR) and Naïve Bayes (NB). These two methods were applied on the 13 attributes and 1 target feature of Cleveland dataset and 10 features of Hungarian dataset from UCI repository. To compare the methods, six performance metrics were calculated: accuracy (ACC), Recall (RE), precision(P), F1_score (F1score) and the Area Under the ROC curve (AUC). Thus, in order to enhance the accuracy of our models, Synthetic Minority Over-Sampling Technique (SMOTE) is used. Furthermore, the results were compared to the results of previous works from the literature. Afterwards, it is found that applied to the Cleveland dataset the accuracy of LR and NB classifiers reached respectively 90% and 89 % and for the Hungarian dataset, the accuracy of the NB and LR algorithm increased to 85% and 81 % Finally, after applying the SMOTE methods the accuracy was improved to 92% and 90% for the Cleveland dataset and 85% and 82% using the Hungarian dataset.