Early Detection of Heart Failure by Reducing the Time Complexity of the Machine Learning based Predictive Model

Abstract

Heart failure (HF) is a deadliest disease. Unfortunately, the diagnosis of HF is very expensive and need high level of human expertise. Hence, there is a dire need of development of automated methods for diagnosis of HF with high accuracy and lower computational cost. To reduce the time complexity of the machine learning model established for detection of HF, in this paper, a hybrid diagnostic system is developed. The proposed hybrid model hybridizes two models i.e., L1 regularized support vector machine (SVM) and linear discriminant analysis (LDA) model. The SVM model is used to remove redundant features or features which are not relevant while the LDA model is used as classifier to classify the input feature vector as that of healthy or diseased person. The elimination of irrelevant features from features space reduces the time complexity of the predictive model, thus, reduces its training time. Simulation results point out that the developed hybrid system achieves better accuracy than twelve previously reported methods that yielded HF risk prediction accuracies in the range 57.8% to 89.01%. Thus, the proposed hybrid model is effective from both aspects i.e. processing time and classification.