Deep Learning Approaches for Smart Stress Detection

Stress has a big impact on people’s well-being, productivity, and health. Automated stress detection technologies have grown in value as wearable sensors and mobile health tracking have proliferated. In order to categorize stress levels based on behavioral and biometric data, this research investigates the use of five deep learning models: Multilayer Perceptron (MLP), Autoencoder + MLP, 1D Convolutional Neural Network (1D CNN), TabNet, and Long Short-Term Memory (LSTM). Model performance is compared in terms of accuracy, speed, complexity, and interpretability using a dataset with 1,000 samples and 10 characteristics. The findings show that while Autoencoder and CNN models offer strong substitutes for feature-rich inputs, MLP and TabNet models present the most promising balance between simplicity and accuracy. The non-sequential nature of the data caused LSTM to perform poorly.