Mixture of autoregressive modeling orders and its implication on single trial EEG classification

摘要: Two methods for mixing AR features EEG signal classification are proposed.Evolutionary and ensemble learning considered.The results assessed against a set of conventional order estimation methods.The feasibilities investigated using several BCI competition datasets.Adequacy Ensemble-based mixture EA-based fusion shown. Autoregressive (AR) models commonly utilized feature types in Electroencephalogram (EEG) studies due to offering better resolution, smoother spectra being applicable short segments data. Identifying correct AR's modeling is an open challenge. Lower model orders poorly represent the while higher increase noise. Conventional estimating include Akaike Information Criterion (AIC), Bayesian (BIC) Final Prediction Error (FPE). This article assesses hypothesis that appropriate multiple likely true compared any single order. Better spectral representation underlying patterns can utility Brain Computer Interface (BCI) systems by increasing timely & correctly responsiveness such operator's thoughts. mechanisms Evolutionary-based identifying orders. The performance resultant AR-mixtures community including (1) A well-known used suggested literature, (2) approaches (e.g., AIC, BIC FPE), (3) blind originated from range Five datasets III contain 2, 3 4 motor imagery tasks considered assessment. indicate superiority evolutionary-based within all datasets.