Performance Prediction
An AI prototype that predicts whether a student is likely to graduate or drop out, using distributed academic data and a real-time Streamlit UI.
Story behind the project
Education teams need early signals when a student is at risk – not after they have already left. This project explores how distributed academic records can feed a practical prediction workflow.
Problem statement
Student outcome data often lives in separate stores. Without integration and a usable prediction interface, institutions struggle to spot dropout risk early enough to intervene.
What was built
The system merges graduate and dropout records from separate SQLite nodes, cleans and scales features, trains a PyTorch neural network, and serves live predictions with confidence scores in Streamlit.
- Integrates graduates.db and dropouts.db sources
- Encoding, scaling, and model persistence (model.pt, scaler.pkl)
- PyTorch neural network for graduate vs dropout classification
- Streamlit UI with real-time prediction and probability display
Tech used
Core tools and decisions that shaped the implementation.
Source & next steps
Full code, setup notes, and ongoing work live on GitHub. Open the repository to review architecture, run locally, or fork the approach.