Inspiration

We became interested in brain-computer interfaces because electroencephalography allows us to record and study brain activity linked to actual movement.. Using brain-computer interfaces can be hard. It takes time to set up equipment understand neuroscience basics work with signal processing and use machine learning models. That’s why we created NeuroSignal AI. We wanted to build a tool that helps people learn the BCI process by working with real EEG recordings. They can see how machine learning models classify motor-imagery data. NeuroSignal AI is a system for analyzing EEG files and showing how classification works

What it does

NeuroSignal AI is a software tool that takes EEG recordings of motor-imagery tasks and runs them through a classification pipeline. It uses bandpass filtering between 8 and 30 Hz applies Common Spatial Pattern to extract features and classifies the data using Linear Discriminant Analysis. Users can look at the raw EEG data. See the output from each step in the pipeline. We also built a BCI experiment and an AI Mentor feature. This mentor explains the science and methods behind each stage.We tested our system using motor-imagery EEG data recorded from our heads with a Muse 2 EEG headset. Our model achieved 87% accuracy and 86% F1-score with LDA. That’s better than the baseline model, which used CSP with Logistic Regression. The baseline had 82% accuracy and 81% F1-score.

How we built

To build the system we first collected motor-imagery EEG recordings. Designed a signal processing pipeline. We applied CSP to the filtered EEG data. Used the extracted features to train a classifier. After finishing the prototype we turned it into an app. We built a user interface that shows the EEG signal the features after training and the classifier weights. We used Logistic Regression as a baseline. Created a mentor interface to explain our approach.

Challenges we ran into

Our project faced times during development. Two days before the deadline our system crashed. We lost the recent version of our code. Recovering it took time. Slowed us down. We had to skip some improvements. Another challenge was that our system didn’t work well across people. EEG patterns vary a lot from one person to another. That made it hard to use a classifier. So we had to change how we trained the model and redesigned the interface. This taught us how important it is to manage data process signals correctly and test models properly.

Accomplishments that we're proud of

“We’re proud that we didn’t stop at training an EEG classifier. We built an entire interactive platform that connects neural signals, machine learning, visualization, and AI-powered explanation—making BCI technology something users can actually explore and understand.”

What we learned

We learned that working with EEG data is more difficult than working with standard machine learning datasets. You have to think about preprocessing methods and how they affect model performance. We also learned how vital it is to split training and test data especially when dealing with data from multiple subject.

What's next for NeuroSignal.AI

We want to make NeuroSignal AI work with data, from subjects and improve model accuracy. We also want to move from recorded data to real-time analysis and run the system on EEG hardware.. We hope to explore new machine learning models and keep improving the interactive BCI experiment.

Built With

Share this project:

Updates

Submission history