CHOSUN

A Study on Design of AI Algorithms for Classification of Mental Stress Based on Electrocardiogram

Metadata Downloads
Author(s)
강민구
Issued Date
2021
Keyword
ECG, CNN, LSTM, SVM, NB
Abstract
Recently, owing to the improvement of the living environment and economic growth, interest in health has increased. Moreover, both physical and mental health management is recognized as an important part of health. In this study, an AI algorithm for classifying mental stress signals using electrocardiogram-based convolutional neural network (CNN)-long short-term memory (LSTM) and support vector machine (SVM)-naive Bayes (NB) models was presented. The R-S peak and R-R interval were extracted from the ECG signals during stress or resting state. To maximize the performance of the stress signal classification algorithm, parameters extracted from ECG signals were applied to the SVM-NB and CNN-LSTM models, and the accuracy of the stress signal classification model was improved by increasing the number of training data through a spectrogram applied with Fourier transform. Subsequently, the performance of the algorithm for each epoch was illustrated in the time-frequency domain and the stress signal classification error. The results of the stress classification algorithm showed that the accuracies of the confusion matrix, receiver operating characteristic (ROC) curve, and precision-recall (PR) curve were 98.3%, 98.12%, and 97.6%, respectively. Therefore, the proposed algorithm can help in the health management of modern people who suffer from mental stress.|최근 경제 성장에 의한 생활환경의 개선으로 건강에 대한 관심이 높아지면서 신체적 건강관리뿐만 아니라 정신적 건강관리 또한 중요한 부분으로 인식되고 있다. 본 연구는 심전도 기반 Convolutional Neural Network (CNN) - Long Short Term Memory (LSTM), Support Vector Machine (SVM) - Naive Bayes (NB) 모델을 이용하여 정신적 스트레스 신호를 분류하는 앙상블 알고리즘을 제시했다. 스트레스를 받았거나 혹은 휴식 상태일 때의 심전도 신호로부터 R-S Peak, R-R Interval를 추출했다. 스트레스 신호 분류 알고리즘의 성능을 극대화하기 위하여 심전도 신호로부터 추출된 파라미터들을 SVM-NB, CNN-LSTM 모델에 적용하였고, 푸리에 변환을 응용한 Spectrogram을 통해 Training 데이터 개수를 증가시켜 스트레스 신호의 분류 모델에 대한 정확도를 향상시켰다. 그 후 Time-Frequency 영역에서 Epoch별 알고리즘의 성능을 나타내었고 스트레스 신호 분류 오차율을 계산했다. Confusion Matrix, Receiver Operating Characteristic (ROC) Curve, Precision-Recall (PR) Curve를 통해 스트레스 분류 알고리즘의 성능을 평가한 결과, 각각 98.3%, 98.12%, 97.6%의 정확도를 나타내었다. 이를 통해 정신적 스트레스에 시달리는 현대인들의 건강관리에 도움을 줄 수 있다.
Alternative Title
정신적 스트레스 신호 분류를 위한 심전도 기반 AI 알고리즘 설계에 관한 연구
Alternative Author(s)
Mingu Kang
Affiliation
조선대학교 일반대학원
Department
일반대학원 IT융합학과
Advisor
김윤태
Awarded Date
2022-02
Table Of Contents
Table of Contents i
List of Figures iii
List of Table v
Acronyms vi
Abstract(Korean) vii

I. Introduction 1
1.1. Research Background 1
1.1.1. Definition and type of stress 1
1.1.2. Definition of ECG signal 2

II. Classification of stress signals using bio-signals 3
2.1. Previous studies 3
2.1.1. The definition of deep learning 3
2.1.2. Database of ECG signals 6
2.1.3. Removing noise and extracting feature points using ECG signals 7

III. Classification of mental stress signals using AI algorithms 9
3.1. Classification of mental stress signals using CNN-LSTM 9
3.1.1. Add number of training data using spectrogram 9
3.1.2. Design a mental stress classification algorithms using CNN-LSTM 10
3.2. Classification of mental stress using SVM-NB 14
3.2.1. Data classification using SVM-NB 14

IV. Experiment result 20
4.1. Stress classification performance evaluation using CNN-LSTM 20
4.2. Stress classification performance evaluation using SVM-NB 28

V. Conclusion 31

References 33

List of Publications 37

Abstract(English) 41
Degree
Master
Publisher
조선대학교 대학원
Citation
강민구. (2021). A Study on Design of AI Algorithms for Classification of Mental Stress Based on Electrocardiogram.
Type
Dissertation
URI
https://oak.chosun.ac.kr/handle/2020.oak/17241
http://chosun.dcollection.net/common/orgView/200000588708
Appears in Collections:
General Graduate School > 3. Theses(Master)
Authorize & License
  • AuthorizeOpen
  • Embargo2022-02-25
Files in This Item:

Items in Repository are protected by copyright, with all rights reserved, unless otherwise indicated.