Attention Aware Deep Learning Approaches for an Efficient Stress Classification Model.

Area of Science:

• Artificial Intelligence
• Computational Linguistics
• Health Informatics

Background:

• Stress significantly impacts modern societies, affecting daily activities and contributing to various diseases.
• Accurate stress measurement is crucial for public health initiatives and improving quality of life.
• Existing stress monitoring systems require advanced classification techniques to interpret physiological and emotional responses.

Purpose of the Study:

• To propose a novel deep learning approach for an accurate stress classification system.
• To develop an Enhanced Long Short-Term Memory (E-LSTM) model integrated with a feature attention mechanism for stress polarity determination.
• To enhance stress detection and prediction capabilities within stress monitoring systems.

Main Methods:

• Utilized a novel deep learning approach, Enhanced Long Short-Term Memory (E-LSTM) with a feature attention mechanism.
• Employed sequential modeling and word-feature seizing for stress classification.
• Evaluated the model using health-related stress data from the Korea National Health and Nutrition Examination Survey (KNHANES VI).

Main Results:

• The E-LSTM model with feature attention achieved an accuracy of 75.54%, precision of 74.26%, recall of 72.99%, and F1-score of 74.58%.
• Demonstrated superior performance in stress detection classification compared to traditional methods like Naïve Bayesian, SVM, Deep Belief Network, and standard LSTM.
• The feature attention mechanism effectively identified complex relationships and extracted relevant keywords for classification.

Conclusions:

• The proposed feature attention mechanism-based E-LSTM approach is efficient for accurately classifying stress.
• This method shows significant potential for enhancing stress monitoring and prediction systems.
• The study highlights the effectiveness of deep learning, particularly E-LSTM with attention, in analyzing complex health-related stress data.