AE-HGNN: attention-enhanced hypergraph neural networks for interpretable stress prediction through higher-order...

Area of Science:

• Computational neuroscience
• Machine learning applications in healthcare
• Wearable technology and digital health

Background:

• Stress significantly impacts mental and physical health, contributing to anxiety, depression, and cardiovascular issues.
• Traditional machine learning models (SVM, Decision Trees, DNNs) face limitations in predicting stress due to difficulties in capturing high-order interactions and handling noisy data.

Purpose of the Study:

• To propose a novel Stress Prediction System using Attention Enhanced Hypergraph Neural Networks (AE-HGNN).
• To model higher-order relationships among environmental and behavioral stress indicators for improved prediction accuracy.

Main Methods:

• Development of an Attention Enhanced Hypergraph Neural Network (AE-HGNN) architecture.
• Leveraging hyperedges to represent multi-node dependencies, going beyond pairwise relationships.
• Incorporating an attention mechanism to assign weights to individual stress factors.

Main Results:

• The AE-HGNN model achieved a test accuracy of 99.75% (5-fold cross-validated mean: 98.44% ± 0.56%).
• Optimized hyperparameters included a learning rate of 0.01, hidden dimension of 128, and epoch size of 150.
• Demonstrated superior performance compared to Random Forest (87.3%), SVM (82.8%), and standard DNNs (90.7%).

Conclusions:

• The AE-HGNN model effectively captures complex, higher-order dependencies among stress indicators.
• The proposed system offers a robust and highly accurate method for stress classification.
• Confirms potential for non-invasive, real-time stress monitoring using wearable devices and mobile health platforms.