Classification of Diabetes and Cardiac Arrhythmia using Deep Learning

Abstract

Deep Learning (DL) is a research area that has ourished signi cantly in the recent years and has shown remarkable potential for arti cial intelligence in the eld of medical applications. The reasons for success are the ability of DL algorithms to model high-level abstractions in the data by using automatic feature extraction property as well as signi cant amount of medical data that is available for training these algorithms. DL algorithms can learn features from a large volume of healthcare data, and then use the procured insights to assist clinical practice. We have implement DL algorithm for the classi cation of two diseases in the medical domain: Diabetes and Cardiac Arrhythmia. Diabetes is often considered as one of the world's major health problems according to the World Health Organization. Recent surveys indicate that there is an increase in the number of diabetic patients resulting in the increase in serious complications such as heart attacks and deaths. This thesis presents a Multi-Layer Feed Forward Neural Networks (MLFNN) for the classi cation of diabetes on publicly available Pima Indian Diabetes (PID) dataset. A series of experiments are conducted on this dataset with variation in learning algorithms, activation units, techniques to handle missing data and their impact on classi cation accuracy have been discussed. Finally, the results are compared with other machine learning algorithms like Na ve Bayes, Random Forest, and Logistic Regression. The achieved classi cation accuracy by MLFNN (82.5%) is the best of all the other classi ers. The term arrhythmia refers to any variation in the usual sequence of the heartbeat. There are many types of cardiac arrhythmia ranging in severity, including Premature Atrial Contractions (PACs), Atrial Fibrillation, and Premature Ventricular Contractions (PVCs). This thesis focuses on the use of DL algorithms: Convolutional Neural Network (CNN) and Long Short- Term Memory (LSTM) to classify arrhythmia with minimum possible data pre-processing on MIT-BIH Arrhythmia Database (MIT dataset). Furthermore, we study the in uence of di erent hyperparameters like L2 regularization and number of epochs on the classi cation accuracy of LSTM. We achieved a classi cation accuracy of 99.19% and 98.40% with CNN and LSTM models respectively. From our research, we believe that CNN model can assist the doctors in the classi cation of arrhythmia.