Using distances to classify recordings of young and elderly subjects

To the large number of diverse methods for the computation of heart rate variability and the common classification algorithms we add a new one, based on the distances between heart rate signals. Given two groups of subjects and a test subject, which we want to categorize into one of the two groups, we compute the average distance of the test subject from each group and then classify the subject based on these two distances. In this algorithm, we do not use distances between groups of features, but distances between the signals themselves. The surprising result is that the signals are not categorized simply based on the minimum average distance from each group. We noticed that distances between signals of elderly subjects were smaller that those between young subjects. Distances between young and elderly subjects were somewhere in the middle. Three distance metrics are used: the Euclidean distance, the Manhattan distance and dynamic time warping.