Hybrid Gradient Boost based Heart Failure Prediction System

Abstract

Heart diseases are prevalent in today's world due to many factors like lipid disorder (hypercholesterolemia), corpulence (obesity), increase in triglycerides levels (lipids obtained from esterification fatty acids to glycerol), hypertension, etc. It is estimated that nearly 18 million lives are affected yearly due to various heart diseases. Early detection of such diseases could help save several lives. In the proposed system, heart failure prediction is estimated using the combination of Gradient boost detection and decision trees. The parallel handling approach is used for feature processing to speed up the results and for optimal performance. The generation and discrimination approach are used to verify the outcomes concerning other algorithms and pseudo-codes. This paper uses the data file from the University of California, Irvine Intelligent Systems Repository to test the results. It is observed from several experiments that it provides optimal performance compared to the remaining predictors in the context of f1 score, recall, and accuracy. The ROC curve of Gradient Boost provides a higher deviation for low false positives. The Gradient Boost shows a 0.919 ROC value and 92 % of accuracy with an F1 score of 0.928 and a recall of 0.934.