fNIRS Signals Classification with Ensemble Learning and Adaptive Neuro-Fuzzy Inference System

Abstract

Brain-Computer-Interface systems were invented in the last decade to record brain signals and then control a system that behaves and conveys with a biosignal recording device and the brain. Its major objective is to aid individuals who suffer from behavioral infirmity. The focus of this research is to analyze the cortical surface of the brain's hemodynamic response using functional near-infrared spectroscopy signals (fNIRS). It is utilized in a variety of cognitive neuroscience and behavioral rehabilitation treatments. Additionally, it was applied to classify thirty participants who volunteered to do a task divided into three classes. The primary task is to classify multi-class fNIRS signals using various classification methods and then compare the results. We utilized classification methods for each of the 30 subjects, followed by the voting and stacking procedures as part of an ensemble learning method. The averaged results for all subjects reached 64.813 percent, while ensemble learning using the voting method reached 66.416 percent. Following that, ensemble learning using the stacking method combined with the ANFIS kernel reached 60.6616 percent. Finally, the findings suggest that it may improve accuracy and reduce standard deviation depending on the Ensemble Learning approach used. It asserts that when the variance of the predictions was reduced, the classification model produced better results.