An Enhanced Probabilistic LDA for Multi-Class Brain Computer Interface
Figure 4
Decision boundaries and feature vector distribution of training sets derived from one of the 5×10-fold cross validation processes for simulated datasets.
The green curve denotes the theoretical boundary, and the boundary curves for BLDA and EBLDA are in red and blue respectively. The blue circles and the red crosses represent the training samples of classes 1 and 2 respectively. (a) training set contains 50 samples; (b) training set has 100 samples; (c) training set has 150 samples; (d) training set has 200 samples. The size of the test set consists of 100 samples for (a), (b), (c) and (d), and EBLDA will select the samples with high probability from these 100 samples to enlarge the training set. The shift of decision boundary between BLDA and EBLDA was due to the combination of reliable samples with high probability in EBLDA.