Stepwise and Quadratic Discriminant Analysis of P300 signals for Controlling a Robot

Brain Computer Interfaces facilitate people with disabilities to lead a better quality of life, one such useful application being the utilization of P300 signals to control the movements of a robot/wheelchair. In this study, feature extraction for P300 signal detection is carried out using discrete wavelet transform (DWT) according to an existing method after signals are averaged over six stimulation trials. Further different combination of feature selection and classification methods have been tested to obtain increased performances for a real-time system. Fisher's criterion and stepwise discriminant analysis (SWDA) were compared for feature selection methods and linear and quadratic discriminant analysis (QDA) and support vector machine (SVM) were explored as classifiers to test out the accuracies of the different feature selection models and a comparison was made between the three classifier models. The combination of SWDA and QDA as a feature selection and classification method after averaging event related potentials (ERPs) over six trials resulted in average accuracies of 96.75 percent across 4 subjects complying to the constraints of a real-time system for movement of a robot in terms of performance and timing.