Analyzing and identifying predictable time range for stress prediction based on chaos theory and deep learning

Area of Science:

• Computational psychology
• Artificial intelligence
• Nonlinear dynamics

Background:

• Stress is a global health concern with significant consequences.
• Predicting stress allows for timely interventions and management.
• Human psychological states exhibit chaotic dynamics, posing prediction challenges.

Purpose of the Study:

• To develop an advanced stress prediction model by integrating deep learning and chaos theory.
• To leverage chaos theory to uncover underlying patterns in stress dynamics.
• To improve the accuracy and timeliness of stress prediction.

Main Methods:

• Embedding stress sequences into a high-dimensional phase space using chaos theory.
• Applying a two-layer (dimension and temporal) attention mechanism in deep learning to model nonlinear stress dynamics.
• Validating the model on the Tesserae dataset for 2-label and 3-label stress prediction.

Main Results:

• The proposed integrated method significantly outperforms pure deep learning and chaos theory methods.
• Stress prediction accuracy improved by over 2% compared to deep learning and 8% compared to the chaos method.
• The model effectively identifies predictable time ranges within stress sequences.

Conclusions:

• The integration of deep learning and chaos theory offers a powerful approach for stress prediction.
• This hybrid model enhances prediction accuracy and provides insights into stress system dynamics.
• Further research can explore additional improvements and applications of this methodology.