DeepNeuroRest: Deep Learning-Assisted Stress Detection in Sleep with Smart Web App Insights

This study conducted a comprehensive analysis of stress level prediction using both machine learning and deep learning models. Phase 1 involved training nine machine learning models, including Decision Tree, Random Forest, Support Vector Machine, k-Nearest Neighbours, Naive Bayes, CatBoost, eXtreme Gradient Boosting, Gradient Boosting, and Extra Trees, alongside five deep learning models: Artificial Neural Networks, Long Short-Term Memory, Radial Basis Function, Gated Recurrent Unit, and Multi-Layer Perceptron. Initially, these models achieved perfect testing accuracy, precision, recall, and F1-Score on the original dataset. However, when incorporating fine-tuned Auxiliary Classifier Generative Adversarial Networks (AC-GAN) to enhance the dataset, the best machine learning performance was observed with CatBoost and Extra Trees, each achieving 98.77% accuracy and an F1-Score of 1 for Class 0, 0.99 for Classes 1 and 2, and 0.98 for Classes 3 and 4. The optimal deep learning model was the Multi-Layer Perceptron, which achieved a test accuracy of 97.852% and a minimal loss of 5.438%. Phase 2 involved developing two client applications to analyse stress levels: an email client and a web client. The best deep learning model was utilized for these applications. The email client, developed using Microsoft Azure, Flask Server, and Microsoft Automation, and the web client, built with Wix, Streamlit, and ngrok, both provided users with stress level analysis depicted through a speedometer graph. These implementations show how the models can be used practically for managing and monitoring stress levels in real-world situations.