A Feasible Classification Algorithm for Event-Related Potential (ERP) Based Brain-Computer-Interface (BCI) from IFMBE Scientific Challenge Dataset

摘要

Event-Related Potential (ERP) based Brain-Computer Interfaces (BCI) have been widely investigated as an alternative human computer interaction solution. For people with neurological diseases and severe disabilities like amyotrophic lateral sclerosis (ALS) and stroke, BCI may be their only access method for communication. For people with neurodevelopmental disorders, such as autism spectrum disorder (ASD), BCI is also considered to be a potential rehabilitation and education assistance method. Although these are promising developments, further work is required to optimize current classification and filtering methods to improve reliability and enhance the user experience. The aim of this project is to investigate the four most commonly used classification algorithms: Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), Convolutional Neural Network (CNN) and Long-Short Term Memory (LSTM) with a novel and personalised filter design. Results of the classification tasks in Phase I and Phase II, of the IFMBE Scientific Challenge, show a 73% accuracy for phase I and 67% accuracy for Phase II.