Two-modes Cyclic Biosignal Clustering based on Time Series Analysis

In this paper we introduce an unsupervised learning algorithm which distinguishes two different modes in a cyclic signal. We also present the concept of “mean wave” which averages all signal waves aligned in a notable point (n zero derivative). With that information the signal’s morphology is captured. The clustering mechanism is based on the information collected with the mean wave approach using a k-means algorithm. The algorithm produced is signal-independent, and therefore can be applied to any type of signal providing it is a cyclic signal that has no major changes in the fundamental frequency. To test the effectiveness of the proposed method, we acquired several biosignals (accelerometry, electromyography and blood volume pressure signals) in the context tasks performed by the subjects with two distinct modes in each. The algorithm successfully separates the two modes with 99.2% of efficiency. The fact that this approach doesn’t require any prior information and the preliminary good classification performance makes this algorithm a powerful tool for biosignals analysis and classification.