أنس هادي

A brain-computer interface (BCI) is a direct communication pathway between a human brain and an external device. In other words, a BCI allows users to act on their environment by using only brain activity, without using peripheral nerves and muscles. In BCI there are many paradigms; one of them is P300 which occurs in response to a significant but low-probability event. BCI data is considered to be high in their dimensionalities which reduce the system performance. Feature selection is a dimensionality reduction technique. Feature selection techniques study how to select a subset of features that enhance the performance of the system. The reason behind using feature selection techniques include reducing dimensionality, removing irrelevant and redundant features, reducing the amount of data needed for learning, and improving algorithms’ predictive accuracy. In this thesis, three types of feature selection techniques are compared and applied. These types are filter, wrapper, and hybrid. Fisher score, Determination Coefficient (r2), Regularized Fisher Linear Discriminant (RFLD), and Bayesian Linear Discriminant Analysis (BLDA) were used as evaluation functions. Differential Evolution (DE) optimization technique was used as searching technique. Two datasets were used to evaluate the results. Filter types were the preferred to be selected as feature selection method for P300 based BCI, in particular r2. This is due to the good reduction in dimension 64.8% and low computational cost 6.75ms. The time required for training and testing the classifier was improved by 83.62%.