Evaluation of Bubble Entropy Using Heart Rate Variability

Bubble entropy has established its own place in the research community, representing a new and promising definition of entropy. Based on the work required to order a vector in an embedding space of dimension m, Bubble entropy gives a physical interpretation of what the metric actually computes. In this work, Bubble entropy is evaluated based on its ability to classify RR time series, the time series most commonly considered for entropy-based analysis in the field of biomedical engineering. For this purpose, it is compared with three other definitions of entropy: the most widely used Sample entropy and Approximate entropy, the most relative to Bubble entropy, and also the widely used Permutation entropy. Signals from healthy individuals, in sinus rhythm, are compared with signals from cardiac patients, and machine learning methods are applied to calculate the classification accuracy that each method can achieve. The classifiers chosen are k-Nearest Neighbors, Support Vector Machine, Logistic Regression, and Gaussian Naive Bayes. Feature evaluation methods are also employed to serve as additional measures of effectiveness. Bubble entropy generally manages to achieve better results than Sample entropy, Approximate entropy and Permutation entropy, both in terms of classification accuracy and feature ranking.