Mind Over Matter

Inspiration

We wanted to learn how to work with complex biosignals and thought the scientific problem was interesting. Only one person in our group has had experience with EEG data and the rest of us were here for the exciting challenge.

What it does

It classifies EEG signals into specific body movements from the labeled dataset that we were given. There are four categories in total.

How we built it

Using Jupyter notebooks on Colab, we trained many of our models and evaluated them. We used ElevenLabs and Gemini API for the visualization website that we used to display the results of our project.

Challenges we ran into

Each subject had very diverse signals, a lot of noise coming from EEG. Deciding how to baseline each patient was also a challenge.

Accomplishments that we're proud of

We managed to achieve more than 50% accuracy on multiple models and we made an ensemble model that took the results of two models and equally weighed them to have a more robust prediction.

What we learned

That working with biosignals is extremely hard due to patient's variability. The data is also quite noisy and if we had time we would have used more frequency domain features.

What's next for Mind Over Matter

Implement more signal preprocessing for better data inputs and test more innovative models to tackle this task. We want to see if we could achieve better performance with data augmentation.

Built With

• collab
• csp
• eegnet
• elevenlabs
• gemini
• github
• mlp
• python
• resnet18
• xgboost