Safe-level SMOTE method for handling the class imbalanced problem in electroencephalography dataset of adult anxious state

Abstract

Anxiety disorder is a mental state in which a person experiences excessive worry, fear, nervousness, and apprehension. Measuring brain signals using the electroencephalography (EEG) modality is one of the ways to detect anxiety. However, an imbalanced EEG dataset class distribution among the existing issues with this method degrades the classification performance of the anxiety state. Therefore, the goal of this research is to improve classification performance by balancing the EEG dataset using a Safe-level Synthetic Minority Oversampling Technique (Safe-level SMOTE). In this work, a freely accessible Database for Anxious States based on Psychological stimulation (DASPS) with 14 EEG channels recorded via headset Emotiv Epoc was employed. The raw EEG signals contaminated with noises were filtered with multiple filtration methods before being further processed. The EEG features were extracted in the time domain, frequency domain, and time–frequency domain for model classification. The features model with the most optimal classification performance was then processed using a sampling technique, and a Safe-level SMOTE based nearest neighbor value of 5 before being classified using k-Nearest Neighbor (k-NN), support vector machine (SVM), and decision tree. Finally, the performance of the dataset was validated using k-fold cross-validation and confusion matrix performance metrics as well as recognition of the subject’s anxiety state. The proposed model indicated that the k-NN achieved the maximum accuracy of 89.5% and the highest precision of 89.7% for the dataset with the enhanced class distribution. The performance of the suggested method with Safe-level SMOTE demonstrates its superiority in recognizing anxiety states compared to existing methods without Safe-level SMOTE.