Inspiration ADHD is often underdiagnosed or misdiagnosed because current methods are slow, expensive, or inaccessible. We wondered: what if everyday digital interactions, like how you move your mouse, could give early signals about attention and focus? That’s how Drift was born: to explore how subtle patterns in cursor motion can reflect cognitive health.
⚡ What it does Drift is a browser extension + interactive simulation that:
Tracks mouse movement and clicks in real time.
Extracts behavioral metrics like reaction time, movement variability, click latency, and path efficiency.
Sends these features to a machine learning model trained on ADHD research datasets.
Displays results as a probability score on a gradient scale (Low, Moderate, High ADHD‑like signals).
Visualizes performance metrics through a sleek, cyberpunk‑style interface.
⚠️ Note: Drift is not a diagnostic tool — it’s a research prototype that demonstrates how everyday digital behavior can provide insight into attention and cognitive patterns.
🛠 How we built it Frontend Simulation: HTML, CSS, JavaScript, with a custom click‑to‑target game to capture attention metrics.
Browser Extension: Chrome APIs to record mouse and click data in background browsing sessions.
Backend: Flask API serving a scikit‑learn logistic regression pipeline.
Model: Trained on ADHD mouse‑tracking datasets (Attentive Cursor Dataset + PLOS ADHD Mouse Tracking Study).
Visualization: Dynamic analytics panel + real‑time probability bar.
🚧 Challenges we ran into Noise in mouse data: raw movement data is messy, we had to normalize and smooth features.
Over‑sensitivity: early prototypes always flagged “ADHD” until we redesigned the output as a probability instead of binary.
CORS and API issues: integrating real‑time predictions between a Chrome extension and Flask backend required debugging fetch requests.
🏆 Accomplishments that we’re proud of Built a working end‑to‑end system: Research paper dataset → ML model → API → extension → live visualization.
Introduced explainable metrics (reaction time, path deviation, etc.) so results are transparent, not a black box.
Showcased how healthcare and web tech can merge into accessible, lightweight screening tools.
📚 What we learned Behavioral signals like cursor motion can reflect deeper cognitive states.
The importance of calibration: models must balance sensitivity with minimizing false positives. This is really important when diagnosing anything.
How to build a full pipeline from start to finish. Research Paper dataset → model → frontend visualization in a short sprint.
🔮 What’s next for Drift Expand to track other conditions (stress, fatigue, motor impairments). Maybe use a radar chart to show everything together. Improve accuracy with larger and more diverse datasets.
Refine the simulation game to generate stronger behavioral signals.
Explore integration into digital health apps for non‑invasive early screening.
Log in or sign up for Devpost to join the conversation.