-
-
User interface of the app (Upload CSV)
-
User interface of the app (Upload Manual Values)
Inspiration Parkinson’s disease affects millions worldwide, often starting with subtle tremors that are hard to detect early. I was inspired to create a tool that helps patients and doctors spot tremors quickly and accurately, empowering early intervention and better care.
What I Learned Through this project, I deepened my knowledge of: Signal processing to extract meaningful features from sensor data. Machine learning models for classification, especially Random Forests. Building interactive web applications that combine AI with user-friendly interfaces. I also learned the importance of data quality, feature consistency, and deploying models for real-world use.
How I Built It The web app takes CSV sensor data as input, extracts key features, and predicts tremor types using a trained AI model. The process includes: -Data preprocessing, cleaning and normalizing sensor readings. -Feature extraction , time-domain and frequency-domain metrics like mean, variance, FFT peaks, and energy. -Machine Learning , trained a Random Forest classifier to detect Rest, Postural, and Kinetic tremors. -Web interface , simple, interactive UI to upload data, get predictions, and visualize results.
Challenges Faced Ensuring feature names in the input matched the model’s expected features. Handling missing or inconsistent sensor data. Optimizing model accuracy while keeping the app fast and responsive.
Built With
Languages & Frameworks: Python, Flask, Pandas, Scikit-learn, Google Collab Deployment: Hugging face
Other Tools: Joblib for model saving, CSV for input data
Built With
- collab
- huggingface
- numpy
- python
- scikit-learn
Log in or sign up for Devpost to join the conversation.