Support Vector Machine based Stress Detection System to manage COVID-19 pandemic related stress from ECG signal

Abstract

This study represents a detailed investigation of
induced stress detection in humans using Support Vector
Machine algorithms. Proper detection of stress can prevent many
psychological and physiological problems like the occurrence of
major depression disorder (MDD), stress-induced cardiac
rhythm abnormalities, or arrhythmia. Stress induced due to
COVID -19 pandemic can make the situation worse for the
cardiac patients and cause different abnormalities in the normal
people due to lockdown condition. Therefore, an ECG based
technique is proposed in this paper where the ECG can be
recorded for the available handheld/portable devices which are
now common to many countries where people can take ECG by
their own in their houses and get preliminary information about
their cardiac health. From ECG, we can derive RR interval, QT
interval, and EDR (ECG derived Respiration) for developing the
model for stress detection also. To validate the proposed model,
an open-access database named "drivedb” available at Physionet
(physionet.org) was used as the training dataset. After verifying
several SVM models by changing the ECG length, features, and
SVM Kernel type, the results showed an acceptable level of
accuracy for Fine Gaussian SVM (i.e. 98.3% for 1 min ECG and
93.6 % for 5 min long ECG) with Gaussian Kernel while using all
available features (RR, QT, and EDR). This finding emphasizes
the importance of including ventricular polarization and
respiratory information in stress detection and the possibility of
stress detection from short length data (i.e. from 1 min ECG
data), which will be very useful to detect stress through portable
ECG devices in locked down condition to analyze mental health
condition without visiting the specialist doctor at hospital. This
technique also alarms the cardiac patients from being stressed
too much which might cause severe arrhythmogenesis.