CNN and Arduino based Stress Level Detection System

Aim:

          The aim of this project is to develop an interactive mobile application utilizing on device machine learning and Internet of Things for efficient control of household electronic devices.

Introduction:

            The increasing prevalence of stress-related health issues has highlighted the need for real-time monitoring and early intervention. This project presents a comprehensive system for emotion recognition and stress level detection using a combination of Natural Language Processing (NLP), deep learning techniques, and biometric sensors. The system utilizes an ESP32 as the central controller, enabling seamless communication between the laptop-based emotion detection module and wearable sensors.

           In the emotion detection module, NLP techniques such as TF-IDF vectorization, tokenization, stop word removal, and stemming are applied to process textual data. Ensemble algorithms like Random Forest (RF) and Logistic Regression (LR) are used for sentiment analysis, classifying text data into relevant emotion categories. The system also incorporates deep learning with Transfer Learning using a pre-trained MobileNet model from TensorFlow or Keras. This model is fine-tuned for emotion recognition by replacing its top layers, enhancing the accuracy of emotion classification.

          For stress monitoring, physiological data is collected using Max30105 and LM35 sensors to measure heart rate, SpO2 levels, and body temperature. The combined data is processed to estimate stress levels, and results are uploaded to a Firebase cloud for remote monitoring. This system aims to provide an effective, real-time solution for stress management and emotional well-being, benefiting individuals and caregivers.

Proposed System:

          The proposed system integrates advanced techniques in Natural Language Processing (NLP), deep learning, and biometric sensor data to offer a comprehensive solution for real-time emotion recognition and stress level monitoring. The system uses an ESP32 microcontroller as the central unit, enabling seamless communication between a laptop running emotion detection algorithms and wearable sensors that measure physiological parameters.

         In the emotion detection module, NLP techniques such as TF-IDF vectorization, tokenization, stop word removal, and stemming are applied to process and analyze textual data. This data is then classified using ensemble algorithms like Random Forest and Logistic Regression, identifying emotions such as happiness, sadness, or stress. Additionally, deep learning is incorporated through Transfer Learning, where the pre-trained MobileNet model is fine-tuned for emotion recognition by replacing its top layers for better accuracy.

        Biometric data from the Max30105 sensor (measuring heart rate and SpO2) and the LM35 sensor (monitoring body temperature) is continuously collected to assess the user’s stress levels. This data, along with the emotion recognition results, is uploaded to the Firebase cloud for remote monitoring and analysis by caregivers or healthcare providers.

          The proposed system aims to provide a real-time, data-driven approach to monitor emotional well-being and stress levels, facilitating early intervention and personalized care.