Reading detection based on EEG signal analysis

Using the experimental data acquired from three subjects, an offline analysis of the EEG signals has been performed in order to detect the attentive reading. The experiment involved 10 seconds reading trials alternating with 10 seconds of rest. The experimental data consists in 4 data sets recorded in different conditions, each set including from 182 to 320 trials. Half of these trials are reading trials, in which the subjects had to read a randomly selected text. In order to analyze the data, the signal power in different frequency bands has been used to build the feature vectors. Using their Pearson coefficients, the most relevant feature vectors were selected to be used for classification. These vectors have been classified using a Bayes classifier and a KNN classifier. The best results have been obtained using the Fp1-F3 signal energy in the [1÷2] Hz frequency band [as an uni-dimensional feature vector]. Using a KNN classifier with k = 7, an error probability less than 20% has been obtained on all data sets.