Table 16; MindCheck: Cortisol Detector

Table Number 16

Inspiration

Cortisol and stress can lead the chronic long-term issues. It is important to be able to quickly detect your current cortisol levels and get advice to address it.

What it does

This project implements an interface that prompts a camera in order to take a picture. Using a Convolutional Neural Network (CNN), the program detects the stress level of the user. Taking into account time, weather, and facial expressions, the program rates the cortisol level of the user. Then, the user is given further advice and an assistant chat-bot in order to reduce these cortisol levels.

How we built it

ML model: Trained using the FER2013 database and used a handmade CNN model from the PyTorch library. Interface: Implemented with Streamlit & OpenWeatherMap AI Chatbot: Implemented with uAgent

Challenges we ran into

Low success rate: Due to lack of processing power and time, the models initially had low success rate.

Accomplishments that we're proud of

ML Models: Able to implement ML models as a backend within the 24-hour time restriction. Intractable interface: Able to implement an interface with Streamlit AI implementation: Added a Chatbot that enables help to be given.

What we learned

• More experience with models
• More experience with Streamlit
• More experience with uAgent
• More experience working in a team
• Utilizing Git in an active environment

What's next for Cortisol Detector

• Refactor and reorganize the files to enable further features.
• Further training of models to increase success rate
• Tweaks to factors and calculations for more accurate cortisol reads

Built With

• fer2013
• python
• pytorch
• pytorchvision
• streamlit
• uagent