Heart disease prediction using input space partitioning and deep learning

Abstract

Advancements in Medical Instrumentation, AI and Machine newlineLearning have together resulted in machines complementing experts in newlinemedical decision making. Medical diagnosis is an important step in this newlinechain. Traditionally treatments and diagnosis were being used with newlinelimited information from patients, but at the advent of computers newlinecapturing and storing data have revolutionized the way that diagnosis newlineand treatment procedures are practiced recently. Storing patient s newlinerecords in digital format has effectively supported healthcare providers, newlinedoctors and patients to share critical information. Using digital formats newlineto store medical records has led to generation of mammoth medical data. newlineThe techniques such as electronic health records, body area networks are newlineemerged to continuously monitor and diagnose patient s health newlineconditions, through the projection of medical sensors and wearable newlinedevices across human bodies. Since the data generated from the body newlinearea networks are continuous and tremendous in volume, various newlinemachine learning algorithms have been applied for the task of prediction newlineof diseases based on relevant medical attributes of a patient. Recently, newlineinnumerable chronic diseases are spreading widely throughout the entire newlineworld, noticed both in developing and developed countries. Among newlinethose hereditary diseases, Diabetics and heart diseases affect human newlinewell-being at a young stage. In this research work a study of existing newlinemachine learning approaches are made for Diabetes, Heart diseases and newlineArrhythmia as a particular case. In phase 1, Decision tree and K-Nearest newlineneighbor algorithms are used for the prediction of Diabetes. In phase 2 newlineHeart disease prediction using Logistic Regression, Naive Bayes, newlineSupport Vector Machine, K-Nearest neighbor, Decision Tree, Random newlinex newlineForest, XGBoost and ANN is studied. An Ensembled model using newlineLogistic regression, Random forest and Naïve Bayes approaches is newlineproposed. Problem of improving the accuracy of the heart disease newlineprediction using input space partitioning has been investigated.