Biosensor-based Wearable Device for Real-Time Stress Estimation

Sammanfattning

Stress is a common part of everyday life, but when it becomes frequent or longlasting it can affect a person’s health, focus, and daily performance. A challenge is that people may not always notice when their stress level is increasing, which creates a need for systems that can monitor stress in a simple and continuous way. Wearable devices can support this by collecting physiological signals from the body and using them to estimate stress in real time. This project investigates the development of a low-cost wearable prototype for real-time stress estimation. The system uses an ESP32 XIAO S3 microcontroller with sensors for GSR, skin temperature, optical pulse-related measurements, and acceleration. Sensor data is transmitted to a laptop using Bluetooth Low Energy, stored in CSV format, and processed in Python for feature extraction and machine learning evaluation. The project compares SVM, Random Forest, and LSTM models, and also tests whether a model trained on the WESADdataset can be applied to prototype data. The results show that the full data collection and machine learning pipeline works, but the current prototype dataset is too small and variable for reliable stress classification. The best prototype-specific model was LSTM, with an F1-score of 0.250. The work shows that the system is technically possible to build, but further improvements are needed for future wearable stress estimation systems.