Estimation of New York Heart Association class in heart failure patients based on machine learning techniques

The aim of this work is to present an automated method for the early identification of New York Heart Association (NYHA) class change in patients with heart failure using classification techniques. The proposed method consists of three main steps: a) data processing, b) feature selection, and c) classification. The estimation of the severity of heart failure in terms of NYHA class is addressed as two, three and, for the first time, as four class classification problem. Eleven classifiers are employed and combined with resampling techniques. The proposed method is evaluated on a dataset of 378 patients, through a 10-fold-cross-validation approach. The highest detection accuracy is 97, 87 and 67% for the two, three and the four class classification problem, respectively.