Optimized EEG Classification Accuracy of Motor-Imaginary Motions using Genetic Algorithms

Electroencephalography (EEG) based brain-computer interface (BCI) systems are rapidly being investigated for use in biomedical applications due in part to them being relatively inexpensive and minimally intrusive. These interfaces are useful for systems that users can input data and get stimuli as outputs. However, there is a significant drawback in that EEG signals can be easily contaminated from other brain activity not pertaining to the task at hand. If too much contamination exists, the signal could be miscategorized and lead to errors or undesirable outputs for the user. In this paper, we are using a genetic algorithm to optimize the 10-20 electrode system for each user to minimize contamination and maximize the classification accuracy of an artificial neural network classifier. This was accomplished by using a neural network on each electrode to classify its signal into one of three imaginary motions. Afterward, all the electrode classifications were fused using a majority voting fusion algorithm. A genetic algorithm was then guided by the results to identify which electrodes will increase accuracy. The average accuracy for the three subjects attained by the genetic algorithm was 84.34%.