A novel physiological feature selection method for emotional stress assessment based on emotional state transition

Zhen Li, Yun Xing, Yao Pi, Mingzhe Jiang, Lejun ZhangFrontiers in Neuroscience 2023 · DOI

Abstract

The connection between emotional states and physical health has attracted widespread attention. The emotional stress assessment can help healthcare professionals figure out the patient's engagement toward the diagnostic plan and optimize the rehabilitation program as feedback. It is of great significance to study the changes of physiological features in the process of emotional change and find out subset of one or several physiological features that can best represent the changes of psychological state in a statistical sense. Previous studies had used the differences in physiological features between discrete emotional states to select feature subsets. However, the emotional state of the human body is continuously changing. The conventional feature selection methods ignored the dynamic process of an individual's emotional stress in real life. Therefore, a dedicated experimental was conducted while three peripheral physiological signals, i.e., ElectroCardioGram (ECG), Galvanic Skin Resistance (GSR), and Blood Volume Pulse (BVP), were continuously acquired. This paper reported a novel feature selection method based on emotional state transition, the experimental results show that the number of physiological features selected by the proposed method in this paper is 13, including 5 features of ECG, 4 features of PPG and 4 features of GSR, respectively, which are superior to PCA method and conventional feature selection method based on discrete emotional states in terms of dimension reduction. The classification results show that the accuracy of the proposed method in emotion recognition based on ECG and PPG is higher than the other two methods. These results suggest that the proposed method can serve as a viable alternative to conventional feature selection methods, and emotional state transition deserves more attention to promote the development of stress assessment.

Introduction

Most current emotion recognition systems rely on "discrete state" analysis, comparing physiological data from a period of rest to a period of stress. However, human emotions are a continuous, dynamic flow. Conventional methods often ignore the transition phase, which may contain the most vital information about how an individual's body reacts to psychological pressure. This paper presents a novel approach to physiological feature selection for emotional stress assessment. Instead of comparing static, discrete emotional states (e.g., "neutral" vs. "stressed"), the authors focus on the emotional state transition—the dynamic process of moving from one emotion to another.

By identifying features that change significantly during these transitions, the researchers aim to develop more efficient and accurate models for stress detection using wearable-friendly signals. This approach addresses the need for interpretable models that can function effectively on low-power devices without requiring the massive datasets typical of "black box" deep learning models.

Methods and Materials

Data Acquisition and Sensors

The study involved 85 healthy university students. To capture the physiological response to emotional stimuli, the researchers collected three peripheral physiological signals: ElectroCardioGram (ECG), Galvanic Skin Resistance (GSR), and Blood Volume Pulse (BVP). These signals were acquired continuously using wearable-grade biosensors.

Participant wears three bio-sensors (ECG, GSR, and BVP) to collect physiological signals during the experiment.

Stimulus and Experimental Protocol

To induce specific emotional states, the authors utilized the International Affective Picture System (IAPS), a standardized set of 120 images categorized by valence (pleasantness) and arousal (intensity). The experimental protocol alternated between 3-minute rest periods and 4-minute emotion induction periods, targeting Neutral, Negative/Stress, and Positive states.

The valence and arousal levels of the IAPS stimuli used to induce emotional states.

Feature Selection Pipeline

The core innovation is the feature selection pipeline (Method 3 or M3), which follows three primary steps:

  1. Feature Extraction: 39 initial features were extracted from the raw ECG, GSR, and BVP signals.
  2. Transition Analysis: Instead of comparing stable blocks of time, statistical tests (t-tests or Wilcoxon signed-rank tests) identified features showing significant differences immediately before and after an emotional transition.
  3. Redundancy Filtering: Pearson correlation was used to discard features with a correlation greater than 0.9, ensuring a lean and diverse feature set.

Key Results

The transition-based method (M3) proved superior in both dimensionality reduction and classification accuracy compared to standard PCA (M2) and stable-state methods (M1):

  • Efficiency: The proposed method selected only 13 features (5 ECG, 4 PPG/BVP, and 4 GSR). This is significantly fewer than the 22 features selected by stable-state methods or the 17 dimensions retained by PCA.
  • Accuracy: For 3-class emotion recognition (Neutral, Positive, Negative), M3 achieved a Recall of 0.885 and an F1-score of 0.903 for ECG and BVP signals. In a binary "stress vs. non-stress" task, combining all three signals led to a peak accuracy of 0.991.
  • Signal Sensitivity: Heart rate variability (HRV) and pulse rate variability (PRV) indicators were found to be particularly sensitive during transitions, making ECG and BVP the most effective signals for stress detection.

Discussion and Conclusion

The "aha moment" of this research is that physiological markers are most sensitive when the body is actively adjusting to a new emotional stimulus. By focusing on these transitions, the researchers created a model that is both highly accurate and efficient enough for real-time monitoring on low-power wearable hardware.

While the results are promising, the authors acknowledge the limitations of a controlled laboratory environment. Future work will focus on testing the algorithm in "wild" environments where physical movement and environmental noise are present, and integrating the optimized 13-feature model into wearable devices for real-time healthcare and rehabilitation feedback.

Summary updated

Abstract

The connection between emotional states and physical health has attracted widespread attention. The emotional stress assessment can help healthcare professionals figure out the patient's engagement toward the diagnostic plan and optimize the rehabilitation program as feedback. It is of great significance to study the changes of physiological features in the process of emotional change and find out subset of one or several physiological features that can best represent the changes of psychological state in a statistical sense. Previous studies had used the differences in physiological features between discrete emotional states to select feature subsets. However, the emotional state of the human body is continuously changing. The conventional feature selection methods ignored the dynamic process of an individual's emotional stress in real life. Therefore, a dedicated experimental was conducted while three peripheral physiological signals, i.e., ElectroCardioGram (ECG), Galvanic Skin Resistance (GSR), and Blood Volume Pulse (BVP), were continuously acquired. This paper reported a novel feature selection method based on emotional state transition, the experimental results show that the number of physiological features selected by the proposed method in this paper is 13, including 5 features of ECG, 4 features of PPG and 4 features of GSR, respectively, which are superior to PCA method and conventional feature selection method based on discrete emotional states in terms of dimension reduction. The classification results show that the accuracy of the proposed method in emotion recognition based on ECG and PPG is higher than the other two methods. These results suggest that the proposed method can serve as a viable alternative to conventional feature selection methods, and emotional state transition deserves more attention to promote the development of stress assessment.
BibTeX
@inproceedings{li2023novel,
  author = {Zhen Li and Yun Xing and Yao Pi and Mingzhe Jiang and Lejun Zhang},
  title = {{A novel physiological feature selection method for emotional stress assessment based on emotional state transition}},
  booktitle = {Frontiers in Neuroscience},
  year = {2023},
  doi = {10.3389/fnins.2023.1138091},
}
Figure