Skip to content

Latest commit

 

History

History
110 lines (77 loc) · 5.04 KB

README.md

File metadata and controls

110 lines (77 loc) · 5.04 KB

Stress Sense : Estimation of stress levels using PPG signals

Stress Sense is a health concerned research project for effective diagnosis of mental stress levels. It involves binary classification of mental stress, excluding the stress in rest condition.

INTRODUCTION

  • Stress has become a major concern in modern societies.

  • It can lead to:

    a) Low Productivity

    b) Health Issues

    c) Increased Cardiac Disease risk

  • This project aims to provide a novel approach in estimating stress using photo-plethysmo-graph (PPG) Signals.

  • Why PPG?

    a) Non-Invasive Method

    b) Low Cost

    c) Easy to measure

RELATED WORKS

image

OBJECTIVES

  • Implement EMD on PPG signal.
  • Feature extraction from IMFs.
  • Develop a ML model to estimate stress level
  • Evaluate the performance of classification model

PROPOSED METHODOLOGY

Step 1: Collection of Required Data

  • Utilized publicly available MAUS dataset. (Link - [1])

image

Step 2: Data Preprocessing

  • Data segregation into low and high sections
  • Data merging of all participants.
  • Applied Standard scaling

Step 3: Emperical Mode Decomposition for Feature Extraction

  • Decomposing ppg signals into Intrinsic Mode Functions(IMFs)
  • Extracting statistical features namely, mean, minimum, maximum and skewness from the different number of IMFs (keeping in mind max imfs for all sets of data)

image

Step 4: Training and Classification

  • Kept test size as 30% of the labeled dataset
  • Implemented Random Forest Classifier
  • Implemented Support Vector Machine(SVM) for classification.
  • Performed Regularization

RESULTS & VISUALIZATIONS

Sample 'Low stress' PPG signal with ite IMFs

image

Sample 'High stress' PPG signal with ite IMFs

image

Final Correlation Matrix (Regularized SVM)

image

Final Performance Report (Regularized SVM)

image

Final Confusion Matrix (Regularized SVM)

image

CONCLUSIONS/OUTCOMES

Empowering Stress Estimation with PPG signals->

  • Significance of Mental Stress Assessment: The evaluation of mental stress plays a pivotal role in the realm of human-computer interaction.
  • Innovative Signal Decomposition: Employed the Empirical Mode Decomposition (EMD) technique to decompose PPG signals into Intrinsic Mode Functions (IMFs) and extract relevant statistical features.
  • Machine Learning for Stress Classification: Trained a Support Vector Machines (SVM) classification model on the extracted features to categorize PPG signals into low stress and high stress states.
  • Achieving Accuracy and Precision: Our approach yielded an impressive accuracy rate of 82% and precision rate of 85% in estimating stress levels using PPG signals.
  • Effective Information Capture: Leveraging EMD and statistical features proved to be an effective method for capturing pertinent information related to stress within PPG signals.

FUTURE SCOPE

  • Try to explore new methods
  • Deploying the model
  • Develop an app for direct diagnosis of mental stress level

REFERENCES

  1. https://ieee-dataport.org/open-access/maus
  2. D. Jaiswal, A. Chowdhury, D. Chatterjee, and R. Gavas, “Unobtrusive smart-watch based approach for assessing mental workload,” in Proc. IEEE Region 10th Symp., 2019, pp. 304–309
  3. F. Schaule, J. O. Johanssen, B. Bruegge, and V. Loftness, “Employing consumer wearables to detect office workers’ cognitive load for interruption management,” Proc. ACM Interactive, Mobile, Wearable Ubiquitous Technol., vol. 2, no. 1, pp. 1–20, 2018
  4. D. Ekiz, Y. S. Can, and C. Ersoy, “Long short-term network based unobtrusive perceived workload monitoring with consumer grade smartwatches in the wild,” IEEE Trans. Affect. Comput., p. 1, 2019, doi: 10.1109/TAFFC.2021.3110211
  5. https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=9664394&tag=1
  6. https://www.mql5.com/en/forum/emd-imf
  7. https://www.researchgate.net/Exploring_EEG_Empirical_Mode_Decomposition

TECH STACKS INVOLVED

  • python

RESEARCH TECHNIQUES INVOLVED

  • Symmetric Projection Attractor Reconstruction (SPAR)
  • Cross Wavelet Transform (XWT)
  • Chirp Z Transform (CZT)
  • Heart Rate Variability (HRV)
  • Emperical Mode Decomposition (EMD)

TEAM

Krishna Dubey (ML and Reasearch), Pankaj Kumar Giri (Reasearch)