Clustering Data with Temporal Evolution: Application to Electrophysiological Signals

Abstract

Electrocardiography signals (ECGs) are typically analyzed for medical diagnosis of pathologies and are relatively unexplored as physiological behavioral manifestations. In this work we analyze these signals by employing unsupervised learning methods with the intent of assessing the existence of significant changes of their features related to stress occurring in the performance of a computer-based cognitive task. In the clustering context, this continuous change of the signal means that it is difficult to assign signal samples to clusters such that each cluster corresponds to a differentiated signal state.

We propose a methodology based on clustering algorithms, clustering ensemble methods and evolutionary computation for detection of patterns in data with continuous temporal evolution. The obtained results show the existence of differentiated states in the data sets that represent the ECG signals, thus confirming the adequacy and validity of the proposed methodology in the context of the exploration of these electrophysiological signals for emotional states detection.