EEG Classifier
Reads raw brainwaves and detects schizophrenia at 94.3% accuracy. A deep learning architecture that listens to the brain's geometry and its rhythm at the same time. Peer reviewed, IEEE published.
- 94.3%
- Accuracy
- CNN+LSTM
- Architecture
- IEEE
- Published
The problem and the build
Diagnosing schizophrenia today still leans on clinical interviews. EEG sees the disorder in the signal, but the signal is noisy, high dimensional, and never the same twice. The architecture I designed treats it the way the brain actually behaves: a CNN front end learns the spatial fingerprint across electrodes, an LSTM back end learns how that fingerprint evolves through time, and the two streams fuse before classification. Months of preprocessing, hyperparameter ablation, and cross subject validation later, the model held 94.3% on a benchmark where prior work plateaued in the high 70s. Published in IEEE Xplore. The takeaway that stayed with me: applying ML to physics, biology, and medicine is where the work is genuinely hard and genuinely matters.
Key insight
Most teams threw bigger models at noisier signals. The win was structural. Match the architecture to the physics. Space and time are different problems. Solve them with different layers, then fuse.
Stack
- PyTorch
- MNE
- NumPy
- scikit-learn
More case studies
- SwarmieFounder validation through grounded AI swarm simulation. Paste a pitch, watch 500 AI personas react like a Reddit thread, and walk away with the three real objections before you burn a single user interview.
- LearnAIA free map of all of AI, from linear algebra to agents. 14 fields, 152 steps, 162 problems. Every step is something you read, then something you build. Finish it and a star lights on your map.
- PitchcraftA pipeline that turns a CSV of local businesses into deployed prototype websites, audit PDFs, and growth proposals, ready to send. One operator, a batch of leads, a folder of finished deliverables.