AI Depression Checker

[ennriqe]

Show and tell project type

Hackathon Project Submission

Github Repo Link

https://github.com/Lucab95/ETH-Online-mental-health/blob/devnet/

Video Walkthough Link

https://youtu.be/lJrGpm0942A

Project Description

Users fill out a form to check for depression, securely uploaded to Nillion. A Neural Network provides a blind prediction. Trained on Kaggle's Depression Anxiety Stress Scales (95% accuracy), it predicts 5 classes.

We use the https://www.kaggle.com/datasets/lucasgreenwell/depression-anxiety-stress-scales-responses dataset to train a simple but highly performant NN (95% accuracy) to predict 5 different classes: Normal, Mild, Moderate, Severe or Extremely Severe.

What problems does your project solve? How does it preserve privacy for users?

• Problem 1: People feeling off and want to check whether they might suffer from symptoms of depression.
• Solution 1: Fill in a form and get back an AI prediction of the extent of their depressive symptoms if any.
• Problem 2: Users do not want to download an app that might sell to third parties their data along with the fact that they have or think they have depression.
• Solution 2: Private upload of the input form and blind computation to give the users the prediction without any data spillover.

How does the project use Nillion? Describe and link to any Nada programs

We train a NN using nada ai
https://github.com/Lucab95/ETH-Online-mental-health/blob/devnet/quickstart_complete/client_code/streamlit_app.py

Is there anything else you want to share?

We originally wanted to present this as a builder development bounty for the deadline that just passed. Still, we couldn't publish the whole blind app on testnet since at some point storing any secret was around 8 Nillion.

Optional - Link your project and team members' social handles

https://x.com/BarruecoEnrique
https://x.com/Luca_brglt

Optional - Team ETH Address(es)

• 0x4f6E0faec10CF04B428eC9E34cd88AE04Fba6062
• 0x4e43b435089bF9e7e38E1470E2dF3bA0B560970d