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

Abstract

Listen

Translate article

A physiological signal-based Human-Computer Interaction (HCI) system provides a communication link between human emotional states and external devices. Accurately classifying these signals is vital for effective interaction, which requires extracting and selecting the most discriminative features to differentiate between various emotional states. This paper introduces the SMOTETomek-Correlated Interactive Reinforcement Learning (ST-CIRL) framework for anxiety classification, which leverages meta-descriptive statistics to enhance the state representation in the reinforcement learning process. Firstly, it addresses class imbalance using SMOTETomek and further reduces dimensionality by pruning redundant features. Secondly, the ST-CIRL framework enhances classification accuracy through the collaboration of multiple agents to select the most informative features using Interactive Reinforcement Learning (IRL).
Further, the paper utilizes classifiers, including Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbour (KNN), and Light Gradient Boosting (LGBM) for anxiety classification. Thirdly, the hyperparameters of these machine learning algorithms are tuned using the Optuna approach to enhance model performance. The proposed ST-CIRL framework achieves a maximum accuracy of 95.35\% and an F1-score of 95.49\% using the LightGBM classifier. Furthermore, the results demonstrate that the proposed approach outperforms current state-of-the-art methods. These findings validate the efficacy of the SMOTETomek method and the innovative feature optimization approach, highlighting the potential of reinforcement learning in enhancing HCI systems and expanding its applicability in intelligent system design.