A cluster based model for brain activity data staging

Brain signals allow a fundamental understanding of the mechanisms underpinning behavioral changes. Currently, standard signal staging of brain electrical activity relies on supervised processes. In this work, we propose an unsupervised data segmentation method based on time series complexity and in the cluster-weighted representation. Receiver operating characteristic curves for predicting transitions on simulated and biological datasets suggest that our model distinguishes accordingly between true and false positive transitions. The ad-vantages revealed were as follows: our method was suited for short-length signals; it was adapted to detect changes of rhythmic oscillations; and it showed no restrictive modeling conditions. Moreover, considering a calibration procedure, this method proved to be suitable for different data samples with a single threshold. This hybrid approach saves time and increases the reliability of post data analysis.

Automated summary

This study introduces an unsupervised data segmentation method based on time series complexity and in the cluster-weighted representation, which accurately predicts brain signal changes. The method is suitable for short signals, able to detect changes in rhythmic oscillations, and does not have strict modeling conditions.