User Biometric Identification Methodology via EEG-Based Motor Imagery Signals

Human brain activities—electroencephalogram (EEG) signals—are likely to provide a secure biometric approach for user identification because they are more sensitive, secretive, and difficult to replicate. Many studies have recently focused on identifying and quantifying important frequency patterns in motor imagery (MI), recorded through EEG. However, there is still a lack of an optimal methodology for recognizing users with EEG-based MI. Therefore, we aimed to propose an EEG-MI methodology that utilizes optimized feature extraction methods and classifiers to improve user-aware accuracy. To accomplish this goal, we extracted four features related to MI and compared the accuracies for recognizing users using a support vector machine (SVM) and Gaussian Naïve Bayes (GNB). We then used the half-total error rate (HTER) to determine whether the results were reliable due to an imbalance problem caused by the differences in the data sizes. Thus, we used a common spatial pattern (CSP) to achieve the highest user identification accuracies of 98.97% and 97.47% using SVM and GNB, respectively. All user recognition accuracies are guaranteed by the HTERs, which are below 0.5. However, CSP has the disadvantage of decreasing accuracy on a small dataset scale. Therefore, we proposed and tested a statistical methodology for estimating a minimum dataset scale to ensure CSP performance. We confirm that the used dataset adequately guarantees CSP performance. This study makes a great contribution to the field of information security by presenting an EEG-MI methodology that improves the identification accuracy in human biometrics based on EEG-MI signals.