College Affordability Prediction

This project provides an end-to-end pipeline for predicting net cost after aid and admission probability for U.S. colleges and universities. It includes:
		ETL scripts to ingest and merge data from JSON and National Center for Education Statistics (NCES) Scorecard CSVs (institution-level & program-level debt).
		Model training scripts that fit regression and classification models using scikit-learn.
		Prediction wrappers to load the latest data and models, and make predictions by school ID or name.
		Streamlit app for interactive exploration, allowing users to select a school and override personal academic stats to see admission chances.

⸻
Prerequisites
	Python 3.12


Installation
	1.	Clone the repo

	cd college-affordability


	2.	Create & activate a virtual environment

python3 -m venv .venv
source .venv/bin/activate     
.\.venv\Scripts\activate    

	3.	Install Python dependencies

pip install --upgrade pip
pip install pandas sqlalchemy scikit-learn joblib python-dotenv rapidfuzz streamlit


	4.	(Optional) API dependencies

pip install fastapi uvicorn



Data & Configuration
		Place raw data files under data/raw/:
		schoolInfo.json (base JSON with 39 profile fields)
		Most-Recent-Cohorts-Institution_05192025.csv (institution-level NCES Scorecard)
		Most-Recent-Cohorts-Field-of-Study.csv (program-level debt)
		By default, the SQLite database is db/app.db. No other configuration is needed. If you wish to point to a different database, set DB_URL in a .env file at the project root:

DB_URL=sqlite:///db/app.db



🗄️ ETL Pipeline

Run the ETL script to load and merge all data into the universities table:

python etl/etl.py

Output:

Institution merge: 311/311 rows matched.
FoS merge: 311/311 rows matched.
Loaded 311 rows -> db/app.db (table 'universities')

Model Training

Train both regression and classification models:

python models/train.py

You’ll see metrics printed:

Regression — MAE: 4,009 | RMSE: 4,797 | R²: 0.286
Positive rate for classification: 21.0%
Classification (GPA/SAT/ACT only) — 5-fold CV F1: 0.756 ± 0.121
Saved models → models/

Models are saved to models/cost_regressor.joblib and models/selectivity_classifier.joblib.

 Predictions Wrapper

A helper module models/predict.py provides:
		predict_cost(school_id_or_name) → float
		predict_selectivity(school_id_or_name) → dict(label, probability)

Example usage:

python - <<'PY'
from models.predict import predict_cost, predict_selectivity
print(predict_cost("3471"))                  # by numeric ID
print(predict_selectivity("Rice University")) # by name
PY

 Streamlit App

Launch an interactive UI:

streamlit run app/streamlit_app.py

Features:
	1.	General Predictions: Dropdown to select any school.
	2.	Personalize: Override HS GPA, SAT, and ACT to see personalized admission chance.
	3.	Metrics: Displays net cost after aid and admission probability.

Project Structure

college-affordability/
├─ data/raw/                   # raw data files (JSON + CSVs)
├─ db/app.db                   # SQLite database
├─ etl/etl.py                  # ETL pipeline script
├─ models/
│   ├─ train.py                # train regression & classification
│   ├─ predict.py              # wrapper to load models & predict
│   ├─ accept_rate_regressor.joblib   # saved regression model
│   └─ selectivity_classifier_calibrated.joblib  # saved classifier
└─ app/
    ├─ streamlit_app.py       # Streamlit front-end