An approach to EEG-based gender recognition using entropy measurement methods

Gender carries significant information related to male and female characteristics. Gender classification using physiological measurements are being studied vigorously. An automated analysis of the electroencephalography (EEG) signal is one of many techniques that may help study brain functions. In this study, EEG data were employed to recognize gender through a novel method used to estimate the differences in the status of brain based EEG data of male subjects and data from female subjects by computing different entropy measurements, including fuzzy entropy (FE), sample entropy (SE), approximate entropy (AE) and spectral entropy (PE). The paper presents an improvement of performance for EEG-based gender recognition with signals collected from 28 subjects. The system employs four entropy measurements as the feature extraction algorithm, used six types of classifiers namely K-Nearest Neighbors (KNN), Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), Quadratic Discriminant Analysis (QDA), and Decision Tree (DT). Three ensemble classifiers were employed, including Bagging, Boosting and vote classifier. The classification results showed that by using an FE feature extractor and SVM classifiers, the classification performance achieved an improved classification performance with an accuracy of 0.995 and AUC of 0.995 when compared to LR (accuracy at 0.949 with AUC of 0.983), DT classifiers (accuracy at 0.961 with AUC of 0.963), RF (accuracy at 0.976 with AUC of 0.993), KNN (accuracy at 0.991 with AUC of 0.993), and QDA (accuracy at 0.966 with AUC of 0.993). Using the Boosting and vote method, classification performance was improved further with an accuracy of 0.996 and 0.998, respectively. It was possible to use EEG signals for gender recognition.