Abstract
Using Machine Learning and Deep Learning to predict cognitive tasks from electroencephalography (EEG) signals has been a fast-developing area in Brain-Computer Interfaces (BCI). Yet, one fundamental challenge is that EEG signals are vulnerable to various noises. This paper identifies two types of noise: external noise and internal noise. External noises are caused by subjects’ movement or sensors’ instability, and internal noises result from the subjects’ random mental activities due to the subjects’ mind wandering during the experiment. When the participants conduct other mental activities researchers cannot infer, it will result in data corresponding to’ unknown’ tasks. We pioneer a Human-In-The-Loop (HITL) machine learning model, EEG4Home, to handle both types of noise and increase the accuracy of predicting known tasks. We introduce a plateau threshold to remove external noise and an unknown threshold set to detect unknown tasks to remove internal noise. Both unsupervised (such as K-Means) and supervised (Such as Random Forests, CNN, and RNN) learning algorithms are implemented in this HITL approach. We use the Thinking1 BCI experiments dataset with sixty subjects (available to academic researchers by request). The average prediction accuracy of known tasks has increased from 56.8% to 65.1%. Overall, this EEG4Home model enables researchers or end-users to gain higher prediction accuracy and more interpretable results.