Towards One-Class Pattern Recognition in Brain Activity via Neural Networks

In this paper, we show how one-class recognition of cognitive brain functions across multiple subjects can be performed at a good (above 85%) level of accuracy via appropriate choices of features. This betters the initial work of Hardoon and Manevitz (where such classification was first shown to be possible in principle, with 60% range on the same data) and work of various groups around the world (e.g. Mourao-Miranda (UCL) and Mitchell et al (CMU) which have concentrated on twoclass classification.) The methodologies for one class classification investigated in this paper are the compression neural network (originally due to Cottrell) and developed by many, including Manevitz and Yousef and a version of one-class SVM due to Scholkopf et al. and Yousef and Manevitz. A genetic algorithm is used to help perform the feature selection.