Controlling the False Positive Rate of a Two-State Self-Paced Brain-Computer Interface

A two-state self-paced brain-computer interface (SP-BCI) divides human mental states into two parts: intentional control (IC) and no-control (NC) states. The state during which the user wants to activate the BCI by is called the IC state. False positive rate (FPR) plays a critical role in evaluating a SP-BCI's usability, where FPR refer to the rate incorrectly classifying NC states. Therefore, development of a method which can not only minimize the FPR and also control it to be smaller than any predefined threshold is of vital importance. In this paper, an imbalanced support vector machine (ISVM)-based FPR control scheme is proposed. This method is able to force the FPR of a two-state SP-BCI to achieve the desired FPR, and this effect is independent of subjects and the feature extraction methods used. In this study, the IC state refers to the state during which the subjects perform an instructed motor imagery task, while during the NC state the subjects are instructed to relax. Spectral power and common spatial pattern (CSP) were used to extract features from EEG signals. Results on the IC and NC EEG data from four participants demonstrate the validity of the proposed ISVM-based FPR control scheme in controlling the FPR to be equal to or smaller than any thresholds, including 5% and 0%.