Term of Award

Fall 2020

Degree Name

Master of Science in Applied Engineering (M.S.A.E.)

Document Type and Release Option

Thesis (restricted to Georgia Southern)

Copyright Statement / License for Reuse

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.


Department of Mechanical Engineering

Committee Chair

Biswanath Samanta

Committee Member 1

Junghun Choi

Committee Member 2

Jinki Kim


A study is presented for mental stress classification based on physiological signals acquired using noninvasive wearable sensors. The study involves data collection using a wearable sensor, signal processing for feature extraction, and classification using the extracted features. Several volunteers participated in the study performing different mental activities in the laboratory with the wearable sensor (Empatica E4) and anonymous data of galvanic skin response (GSR), blood volume pulse (BVP), heart rate (HR), and 3-axis acceleration were transmitted via Bluetooth to a laptop and recorded. The data were annotated with anonymous identifier as Px, x denoting the number, individual activity and perceived stress level expressed by the participant. The signals were preprocessed through filtering and later processed using fast Fourier transform (FFT) and multifractal detrended fluctuation analysis (MFDFA) to extract features that included FFT peaks and Hurst exponents. The extracted features were used as inputs to two unsupervised learning algorithms, namely K-nearest neighbors (KNN) and self-organizing map (SOM) for classification of mental stress at five levels: low (L), medium low (ML), medium (M), medium high (MH), and high (H). The extracted features were used for individual signals (GSR, BVP, HR) as well as all combined, both without and with normalization. The distance measures, Euclidean, Manhattan, and Minkowski were used. The features were also processed through principal component analysis (PCA) to see the feasibility of using principal components in classification. An ensemble classifier based on majority voting was used for final classification. The details of data acquisition protocol, acquired signals, extracted features, and classification results are presented. The results show the effectiveness of the extracted features in classification of mental stress levels with good correlation among the classifiers. The future scope of work would include integrating the developed classification process for detection, alleviation, and reduction of mental stress through noninvasive, friendly intervention of robots with human-robot interaction (HRI).

Research Data and Supplementary Material