A Robust Deep Learning based Prediction System of Heart Disease using a Combination of Five Datasets

Abstract

All across the world, heart disease is regarded as a fatal disease. Heart disease is a condition that affects both men and women equally and may be a major cause of death around the world. Early diagnosis of this condition is critical for everyone in order to reduce mortality rates day by day. Chronic kidney disease dataset, from UCI machine learning library, having 1190 samples with 14 characteristics has been used for this study. To make this research more potent, both Machine learning (ML) and Deep learning (DL) techniques were used to detect the sickness early. The data was normalized by standard scaler for having a class varience issue. We then used three deep learning techniques namely Convolutional Neural Network (CNN), Artificial Neural Network (ANN), and Long Short Term Memory (LSTM) with two other general machine learning approaches such as Decision Tree and Support Vector Machine (SVM). To show a replication study, the overall experiments were done based on the three different random subsets. For the classification measurement, we also employ the ROC and the AUC curves. Several promising outcomes have been achieved. We calculated accuracy, precision, sensitivity, specificity, and F1-score. CNN provided the best results, with an accuracy of 99.16%.