Design of an Augmented Ensemble Heart Failure Prediction Model using Multi Parametric Analysis

Abstract

Diseases like congenital heart disease, coronary artery disease, arrhythmia, cardiomyopathy, and pulmonary stenosis are the major reasons for heart failure. Due to these diseases failure conditions like left-sided failure, right-sided failure, diastolic failure, and systolic failure are most frequently observed. A wide variety of body parameter abnormalities constitute towards these conditions. For instance, factors like patient’s age, anaemia dosage, creatinine and phosphokinase levels, presence of diabetes, ejection fraction of heart, presence of high blood pressure, number of platelets, levels of serum creatinine & serum sodium, smoking frequency, etc. are observed to directly affect normal heart functioning. Efficient analysis of these parameters must be done in order to estimate probability of heart failure, which can assist in better clinical care. Currently researched models like convolutional neural networks (CNNs), random forests (RFs), support vector machines (SVMs) are limited in terms of accuracy, and precision performance when applied on multiple types of patient datasets. In order to improve this performance, the underlying text proposes design of an augmented ensemble learning model that combines RFs, Decision Tree (DT), Gradient Boosting (GB), Linear Regression (LR), artificial neural network (ANN), Naïve Bayes (NB), SVM with radial basis kernel, SVM with linear kernel, k Nearest Neighbour (kNN), long-short-term-memory (LSTM) based recurrent neural network (RNN), and LR with selected features. Performance of these classification models is evaluated individually, and then an entity-pattern-level model selection process is performed. This process assists in selection of the best performing model for a given feature pattern, thereby improving overall efficiency of the system. It is observed that the proposed model achieves an accuracy of 99.5% on different datasets, which is superior when compared to various state-of-the-art techniques & various CNN architectures. Due to such a high performance, the proposed model is observed to be applicable for various clinical purposes.