Feasibility of Machine Learning applied to Poincaré Plot Analysis on Patients with CHF

As an alternative to the traditional methods of analysis in the time and frequency domains regarding heart rate variability, new interest has been concentrated in using a non-linear analysis technique of the beat-beat time series, known as the Poincaré Plot Analysis. The parameters provided by the analysis can be used as input for machine learning algorithms in order to distinguish patients in three classes of congestive heart failure, according to the New York Heart Association. Tree-based algorithms for classification and synthetic minority oversampling technique (SMOTE) for balancing the dataset with artificial data were implemented in Knime analytics platform, reaching an overall accuracy between 75% and 80%, specificity and sensitivity greater than 90% in some classes and F-measures ranging from 68% to 92%. Further investigations could be pursued with bigger datasets and avoiding the use of artificial data to balance the classes.