Detection of stress from stressful environments for visually impaired people (VIP) using EEG band signals

Abstract

This paper proposes a model to detect stress from stressful environments for Visually Impaired People (VIP) using EEG signals; EEG is now broadly used for emotion based recognitions and in Brain Computer Interface (BCI). The World Health Organization (WHO) notes stress as the next significant issue of this era which constantly causes damage over physical and mental health of people all over the world. According to WHO's estimation, visual impairment is found in 285 million people around the world and 80% of visual impairment can be prevented or cured if proper treatment is served. However, Visually Impaired People around the world have a concerning rate of living with stressful environments every day. Thus, these motivated researchers to seek a stress detection model which may be used further for supporting Visually Impaired People and higher research purposes. This model refers to work with EEG Bands and detect stress by extracting Absolute Band Power, Average Band Power, Relative Band Power, Standard Deviation and Spectral Entropy from five EEG bands. Support Vector Machine (SVM), K Nearest Neighbors (KNN), Random Forest and Linear Discriminant Analysis (LDA) are used for classification considering their reliability for Multi-Class Classification. Moreover, Stratified 10 Fold Cross Validation method is implemented to ensure the balanced distribution between multiple classes during train and test split in this model. With this experimental dataset, we achieved the best result using Random Forest Classifier (99%) for every environment where SVM, KNN and LDA could secure more than 89% Classi cation Accuracy, Precision, Recall and F1 Score.