Uncertainty Management by Feature Space Tuning for Single-Trial P300 Detection

Abstract

The P300 is a widely studied event-related potential, which allows non-muscular communication. In P300 induced brain–computer interfacing, one often comes across the challenge of modeling uncertainties due to fluctuations in EEG feature values within a specific session and across several sessions of EEG recordings of a specific subject. The relevance of fuzzy systems in this domain thus cannot be undermined. In this paper, the authors propose (a) an interval type-2 fuzzy classifier for detecting P300 occurrences and (b) a feature tuning algorithm for selection of Autoregressive Yule Parameter features of optimal lag-length corresponding to individual electrodes with an aim to maximize a classifier-oriented performance metric. The classifier performance metric is formulated as a simple objective function tailored to the classifier performance in terms of low uncertainty and high classification accuracy. The relationship between the proposed objective function value and classification accuracy is found to be statistically significant over iterations. The experimental results show that the proposed algorithm achieves an average accuracy of 90.8%.