ST-CIRL: a reinforcement learning-based feature selection approach for enhanced anxiety classification.

Area of Science:

• Physiological signal processing
• Human-Computer Interaction (HCI)
• Machine Learning for Affective Computing

Background:

• Effective Human-Computer Interaction (HCI) relies on accurately interpreting human emotional states from physiological signals.
• Classifying these signals requires robust feature extraction and selection to differentiate emotions.
• Existing methods face challenges with class imbalance and feature redundancy.

Purpose of the Study:

• To introduce the SMOTETomek-Correlated Interactive Reinforcement Learning (ST-CIRL) framework for enhanced anxiety classification.
• To leverage meta-descriptive statistics for improved state representation in reinforcement learning.
• To optimize feature selection and classification performance in HCI systems.

Main Methods:

• Addressing class imbalance with SMOTETomek and reducing dimensionality by pruning redundant features.
• Employing Interactive Reinforcement Learning (IRL) with multi-agent collaboration for informative feature selection.
• Utilizing and tuning classifiers (Random Forest, SVM, KNN, LightGBM) with the Optuna approach.

Main Results:

• The ST-CIRL framework achieved a maximum accuracy of 95.35% and an F1-score of 95.49% using the LightGBM classifier.
• The proposed approach demonstrated superior performance compared to current state-of-the-art methods.
• Validation of SMOTETomek for imbalance handling and feature optimization's effectiveness.

Conclusions:

• The ST-CIRL framework significantly enhances anxiety classification accuracy in HCI systems.
• Reinforcement learning shows strong potential for improving physiological signal-based HCI.
• The developed feature optimization strategy is effective for intelligent system design.