P300 Brain-Computer Interface Based On Methods Addressing Class Imbalance Problem

Brain-compute interface is a set of software and hardware system that people could used to exchange information with external devices. It converts EEG activi-ties collected directly in the human brain into control commands by signal processing methods and classification algorithms, without the need to use a peripheral nerve or muscle tissue. BCI technology provides a new control channel for people to commu-nicate with outside world. It has a wide range of applications in medical rehabilitation, disease diagnosis, traffic aids control and other fields.Among so many type brain signals, P300 is one of the most widely used in BCI systems which is generally acquired by Oddball paradigm. However, we found that the data sets of this kind BCIs always have class imbalance problem. The class imbalance problem may have negative effects on the performance of classification algorithms. Hence, the class imbalance problem can not be ignored in the design of BCI signal processing and classification algorithms. But the exist BCI systems failed to take this problem into account. In order to solve the above problem, this paper proposed a P300 detection model based on methods addressing class imbalance problem.The main works in this paper are as follow:(1)Studying and analysing the class imbalance problem of data sets acquired by P300 Speller.(2)Proving that the class imbalance problem is widely exist in P300 data sets, and further discussing the effects this problem might bring to the P300 BCI system.(3)Proposing a P300 detection model based on methods addressing class imbal-ance problem.(4)Proposing SMOTE based Fisher linear discriminant analysis algorithm in the training process to verify the effectiveness of P300 detection model based on methods addressing class imbalance problem.(5)Continue proposing Random-Under Sampling based ensemble linear support vector machines algorithm in the training process to further validate the effectiveness of P300 detection model based on methods addressing class imbalance problem.