Feature Extraction and Classification of EEG Signals Using Machine Learning Algorithms for Biometric Systems

Abstract

EEG (electroencephalogram) based biometrics systems are used in very high-security areas due to its several advantages over traditional biometric systems. This paper presents an approach for extracting features and classification of EEG signals acquired from users for authentication purposes. The Autoregressive (AR) model with order three features is calculated because the AR model features reveal the signal's intrinsic characteristics. An experiment is performed on many classifiers to classify the extracted features. Classifiers are tested with different kernels and optimizers to accomplish good accuracy for the system. Machine learning algorithms such as support vector machines (SVM), k-nearest neighbor (k-NN), multilayer perceptron (MLP), XGBoost are used as classifiers to classify the signals for authentication. Cross-validation is used for splitting data in the train and test set so that more accurate results were obtained on unseen data. 10-fold cross-validation is used in the proposed work. Obtained results show that mean accuracy values up to 99.7% is achieved; in some trials, accuracy up to 100% is achieved with few classifiers. A comparison table is shown, which compares the accuracy values obtained by different classifiers using different kernels and optimizers.