A web-based toolkit for recording and analysis of physiological signals in response to affective and cognitive stressors

KEMAL BEŞKARDEŞLER

A web-based toolkit for recording and analysis of physiological signals in response to affective and cognitive stressors

Thesis Type: Postgraduate

Institution Of The Thesis: Galatasaray University, Fen Bilimleri Enstitüsü, BİLGİSAYAR MÜHENDİSLİĞİ ANABİLİM DALI, Turkey

Approval Date: 2019

Thesis Language: English

Student: KEMAL BEŞKARDEŞLER

Supervisor: Reis Burak Arslan

Open Archive Collection: AVESIS Open Access Collection

Abstract:

Stress is one of most important problems of humanity related to several disorders like depression, obesity, cardiovascular and diabetic diseases. We may understand what stress changes in human physiology by monitoring physiological signals in human body and may prevent these disorders. Physiological signals can be monitored by electrodermal activity (EDA), electrocardiography (ECG), electroencephalography (EEG), electromyography (EMG), photoplethysmography (PPG) and analysis of changes in the human body like temperature, respiration, iris and pupil parameters. In this study, analysis of existing studies regarding stress detection using physiological signals is performed. For the purpose of producing experimental dataset and comparison with existing results, a website to perform stroop color word (CWT), image and calculation tests are developed as controlled environments. Dataset for analyses are collected with Empatica E3 wristband having EDA, temperature, photoplethysmograph and 3-axis accelerometer sensors, and Bitalino kit with EDA and ECG sensors. Analyses are carried out on test type basis with various classification algorithms and binary/trio classifications of rest, stressed, non-stressed, low-stress and high-stress phases are studied. Best results for stroop test are achieved with 83.09% accuracy for "rest vs stressed" classification. 89.61%, 88.35%, 85.78% and 77.46% accuracies are achieved for symbol digit modalities & math test for "rest vs stressed", "rest vs low stress", "rest vs high stress" and "low vs high stress" classifications, respectively. 85.06% accuracy is achieved in image tests' "rest vs stressed" analyses.