A New Self-Regulated Neuro-Fuzzy Framework for Classification of EEG Signals in Motor Imagery BCI

One of the major problems associated with the motor imagery (MI) electroencephalogram (EEG) based brain–computer interface (BCI) classifications is the informative ambiguities mainly caused by interferences of artifacts and nonstationarities in EEG signals. Other factors containing mislabeling or misleading MI EEG trials might also cause more uncertainties in training datasets that lead to decline in classification performance. This paper proposes a new framework to achieve more efficient classification in multiclass MI EEG-based BCIs. An artifact rejected common spatial pattern (AR-CSP) method is proposed for feature extraction in order to cope with the interferences of artifacts. A self-regulated adaptive resonance theory based neuro-fuzzy classifier that is referred to as “self-regulated supervised Gaussian fuzzy adaptive system Art (SRSG-FasArt)” is introduced to deal with EEG nonstationarities. A metacognitive self-regulatory-based learning algorithm is also proposed to more efficiently deal with the uncertainties. The algorithm captures the training data samples by priority and automatically creates, upgrades, or prunes the fuzzy rules by scanning the knowledge content existing in the data patterns and the created rules. The mechanism improves the generalization capability of the SRSG-FasArt and prevents over-training. The performance of the proposed cooperative framework of AR-CSP and SRSG-FasArt is evaluated using the BCI competition IV dataset 2a. The results indicate more accurate and efficient BCI classification compared to the existing frameworks.