🎓 College Debt Outcome Mismatch Analyzer

Surfacing predatory academic programs using federal data — because students deserve to know what a degree actually costs.

---

📄 View Full Analytical Report (PDF) — 11-page report with all findings, charts, and methodology for non-technical readers

The Problem

Every year, over 2 million Americans enroll in college programs without access to a fundamental piece of information — what graduates from that program actually earn, and how much debt they carry when they leave.

The U.S. Department of Education publishes this data annually through the College Scorecard API. It is free, comprehensive, and covers every accredited institution in America. But it is buried in machine-readable files that no student, parent, or counselor has time to interpret.

This tool does that interpreting for you.

---

What It Does

• 📡 Pulls 6,322 institutions from the U.S. College Scorecard API
• 💰 Computes a Value Score (median earnings ÷ median debt) for 4,500+ schools
• 🤖 Uses Isolation Forest anomaly detection to flag 69 predatory institutions across 4 dimensions simultaneously
• 💬 NL→SQL interface powered by Claude API — ask questions about any school in plain English
• 📊 Interactive Streamlit dashboard with tabbed charts, state drill-down, and sidebar filters
• 📄 One-click PDF report export with full findings, charts, and methodology for non-technical readers

---

Key Findings

#	Finding	Implication
1	Mean value score: 3.22 across 4,500 institutions	Most colleges return decent value
2	46 schools have graduates earning less than they borrowed	Break-even violations — concentrated, not systemic
3	69 institutions flagged as multi-dimensional outliers	Poor across debt, earnings, AND completion simultaneously
4	New Jersey ranks 6th nationally — top 2 value schools both in NJ	Strong local options for Rutgers students
5	School size: r = 0.09 correlation with value score	Brand name is a poor proxy for financial outcomes
6	Flagged schools cluster at $5K–$25K debt	Moderate cost can still produce poor outcomes
7	Puerto Rico has highest flagged concentration	Geographic wage gaps create structural metric bias

---

Dashboard Features

4 interactive tabs:

• 🏫 Worst Value Schools — adjustable slider, red bars = below break-even
• 🗺️ State Breakdown — best/worst states, drill into any state's full school list
• ⚠️ Flagged Institutions — sort by value score, debt, or completion rate
• 📈 Distributions — value score, debt, and earnings histograms with summary stats

Sidebar filters: state selector, flagged-only toggle, value score range slider

Plain English query interface: type any question, Claude converts it to SQL, results display instantly

PDF export: one-click generation of a full 11-page analytical report with all findings and charts

---

Tech Stack

Python · pandas · NumPy · scikit-learn · SQLite · SQLAlchemy
Streamlit · Anthropic Claude API · ReportLab · Matplotlib
U.S. College Scorecard API

---

How to Run

git clone https://github.com/rushil1356/College-debt-analyzer
cd College-debt-analyzer
python -m venv venv
venv\Scripts\activate        # Windows
# source venv/bin/activate   # Mac/Linux
pip install -r requirements.txt

Create a .env file in the project root:

ANTHROPIC_API_KEY=your_claude_api_key_here
COLLEGE_SCORECARD_API_KEY=your_scorecard_api_key_here

Get your free API keys:

• College Scorecard: https://collegescorecard.ed.gov/data/documentation/
• Anthropic Claude: https://console.anthropic.com

Run notebooks in order to build the database:

1. data_pipeline.ipynb         — pulls API data, computes value scores, loads SQLite
2. anomaly_detection.ipynb     — isolation forest, flags predatory institutions

Launch the dashboard:

streamlit run app.py

Generate the PDF report standalone:

python generate_report.py

---

Project Structure

college-debt-analyzer/
├── data_pipeline.ipynb          # API ingestion, value scoring, SQLite load
├── anomaly_detection.ipynb      # Isolation forest, flagged schools, findings
├── app.py                       # Streamlit dashboard + NL→SQL interface
├── generate_report.py           # PDF report generator (ReportLab)
├── requirements.txt             # All dependencies
├── .env                         # API keys — never committed
└── .gitignore

---

Sample Questions (NL→SQL Interface)

• "Which schools in New Jersey have the highest value score?"
• "Show flagged schools with median debt over $30,000"
• "Which states have the most predatory institutions?"
• "List schools with completion rate above 80% and value score above 4"

---

Data Source

U.S. Department of Education — College Scorecard API https://collegescorecard.ed.gov/data/documentation/

---

Author

Rushil Pandya · CS Sophomore · Rutgers University '28