4.3. Classification

In this study, the classifiers of the linear discriminant analysis (LDA) and the support vector machine (SVM) were used. LDA is one of the most commonly used classification algorithms in BCI systems [44,45], as it is a simple but accurate method for the identification of EEG signals. The LDA algorithm determines the optimal axes in terms of classification by increasing the variance between the classes and decreasing the variance within the class [46].

The SVM algorithm is robust in binary classification and is considered one of the most accurate classifiers to detect P300 evoked potentials [25]. The SVM separates the data from two classes by finding a hyperplane with the maximum possible margin [47]. SVM can use different kernel functions, the most used are [48]:

Radial Basis Function, K(xi,xj)=e−‖xi−xj2‖2σ2,σ≠0.

Polynomial, K(x_i , x_j) = (x_i × x_j + 1)^d,  d  > 0.

Sigmoidal, K(x_i , x_j) = tanh(kx_i × x_j  −  δ).

Cauchy, K(xi,xj)=(1+‖x−y‖22σ2)−1,σ≠0.

Logarithmic, K(x_i , x_j) = − log(‖x − y‖^d  +  c),  d  > 0.

In the classification process only 100 s of the signal acquired in the experiment was used (50 s from each group). The feature vector that was obtained was divided into two equal parts: training and testing. Then, the training vector was divided into two groups (P300 and non-P300), which was used to train the classifier.

The test vector was formed with samples of type P300 and non-P300 distributed alternately (10 segments of 5 s). It was used to verify the efficiency of the classifier. The performance was established according to the number of samples correctly classified in the P300 and non-P300 groups with the LDA and SVM classifiers. The kernel functions used in the SVM were linear, quadratic, and radial basis. In the Gaussian radial base kernel function (RBF), a scale factor (sigma) of 1 and a penalty parameter (C) of 1 were used. The complete methodology with the different algorithms used in this study is shown in Figure 4.

Methodology used in the processing of EEG signals.