A deep learning assisted Ant Lion Optimization Model for stress detection and classification using Heart Rate

Abstract

The increasing focus on mental health has amplified interest in stress detection and classification systems. Classifying stress levels using physiological data has shown promise for both traditional machine learning (ML) and deep learning (DL) approaches. However, achieving ultra-high accuracy in multi-class stress classification remains challenging for current methods, particularly when utilizing the SWELL-KW dataset. To address this issue, the present study introduces an innovative DL model assisted by Ant Lion Optimization (ALO) for feature selection. The ALO algorithm enhances the model's overall performance by effectively selecting the most relevant features with rapid convergence, overcoming limitations such as overfitting often observed in traditional approaches like Recursive Feature Elimination (RFE) or Principal Component Analysis (PCA). This study highlights the significance of using comprehensive evaluation metrics beyond the conventional F-measure, accuracy, precision, and recall, emphasizing metrics like Cohen's Kappa, Root Mean Square Error (RMSE), and Matthews Correlation Coefficient (MCC) for a more complete assessment of the model's performance. The proposed framework involves data preprocessing, Ant Lion Optimization for feature selection, and classification using Deep Belief Network (DBN), Multi-Layer Perceptron (MLP), and XGBoost algorithms. This approach offers an effective solution for stress detection, with the potential to outperform existing models in terms of accuracy and adaptability.