Electroencephalogram (EEG) signals are the electrical activities of brain recording caused by the synaptic activations of the brain's neurons which are recorded along the scalp surface. Throughout the years, there have been various researches on the implementation of EEG signals as biometric authentication. This is due to the potential of the human brain that is unique and high resistant to forgery. This research analysed the experiment on the EEG based authentication during the performed familiar and unfamiliar typing tasks. A total of 30 subjects (right-handed) with the age of between 19 to 23 years old were chosen to perform two times typing tasks (familiar and unfamiliar) for 3 minutes with rest in between for 1 minutes. The subjects typed their first and last name followed by typing random names for 3 minutes each. They were required to rest before and after performing the typing tasks for one minute. The EEGOTM sports device (ANT Neuro, Enschede, The Netherlands) with frequency sampling of 512 Hz and 32 channels were used. This research applied Independent Component Analysis to remove eye blinks, notch filter to remove 50 Hz powerline artefacts, and bandpass filter to separate the signals into sub frequency bands such as delta, theta, alpha, beta and gamma. The features were extracted through linear feature extraction (Welch’s method, Burg’s method, Yule–Walk’s method) and non-linear features (Fuzzy Entropy) were also extracted. The extracted features were classified through non-linear classifiers (k-Nearest Neighbour, Random Forest and Ensemble Bagged Tree classifiers) to obtain the performance of the experimental data. The extracted features that obtained high performance accuracy were then hybrid among them based on frequency bands and brain lobes through concatenation. The feature selection techniques such as statistical t-test and Mann-Whitney U test were also applied to the extracted features by reducing the number of input variables which improve the classifier performance. The classification results of the features extracted and hybrid by the classifiers showed an improvement of performance accuracy of familiar and unfamiliar typing tasks. In conclusion, the highest percentage accuracy for typing biometrics by using Burg’s method for frontal and parietal lobes hybrid feature which is 95.94% by using Ensemble Bagged Tree classifier.
