Electroencephalography Based Machine Learning Framework for Anxiety Classification

Anxiety is a psycho-physiological phenomenon related to the mental health of a person. Persistence of anxiety for an extended period of time can manifest into anxiety disorder, which is a root cause of multiple mental health issues. Therefore, accurately detecting anxiety is vital using methods that are automated, efficient and independent of user bias. To this end, we present an experimental study for the classification of human anxiety using electroencephalography (EEG) signals acquired from a commercially available four channel headset. EEG data of 28 participants are acquired for a duration of three minutes. Five different feature groups in time domain are extracted from the acquired EEG signals. Wrapper method of feature selection is applied, which selects features from two feature groups among the five feature groups initially extracted. Classification is performed using logistic regression (LR), random forest (RF), and multilayer perceptron (MLP) classifiers. We have achieved a classification accuracy of 78.5% to classify human anxiety by using the RF classifier. Our proposed scheme outperforms when compared with existing methods of anxiety/stress classification.