🍽️ HungerSight — Predictive Food Insecurity Intelligence Platform

A data analytics portfolio project targeting food insecurity reporting for Central Florida nonprofits — built to demonstrate SQL, Tableau, Python ML, and full-stack data delivery skills.

---

🎯 The Problem

Food banks and hunger relief organizations like U.S. Hunger typically operate reactively — responding to demand spikes after they've already peaked. They also lack a unified view of which partner organizations deliver the best outcomes per dollar spent.

HungerSight solves this by:

• Predicting food insecurity risk by ZIP code before it spikes
• Scoring partner organizations by composite impact efficiency
• Delivering actionable insights to stakeholders through interactive dashboards

---

🌐 Live App Features

🏗️ Project Architecture

HungerSight/
├── etl_pipeline.py          # Automated data ingestion & SQLite database builder
├── ml_model.py              # Random Forest risk prediction model
├── app.py                   # Streamlit interactive web application
├── sql/
│   └── analytical_queries.sql   # 22 analytical SQL queries (CTEs, window functions, views)
├── data/
│   ├── hungersight.db       # SQLite database (6 tables, 6 views)
│   └── models/
│       ├── risk_model.pkl   # Trained Random Forest model
│       └── model_metrics.json
├── tableau_exports/         # CSVs for Tableau Public connection
│   ├── county_profile.csv
│   ├── partner_efficiency.csv
│   ├── zip_risk.csv
│   ├── insecurity_trends.csv
│   └── zip_predictions.csv
└── docs/
    └── data_dictionary.md

---

📊 Data Sources

Source	Data	Access
U.S. Census Bureau	Poverty rate, median income, SNAP participation	Public API
USDA Economic Research Service	Food desert scores, grocery access	Public download
Bureau of Labor Statistics	Monthly unemployment by county	Public API
Feeding America	County-level food insecurity rates (2019–2023)	Public CSV

Note: Data is synthetically generated from real-world distributions to preserve privacy and enable reproducibility. All statistics reflect authentic patterns derived from official sources.

---

⚙️ Setup & Run

1. Clone the repository

git clone https://github.com/YOUR_USERNAME/HungerSight.git
cd HungerSight

2. Install dependencies

pip install -r requirements.txt

3. Build the database

python etl_pipeline.py

4. Train the ML model

python ml_model.py

5. Launch the Streamlit app

streamlit run app.py

---

🤖 Machine Learning Model

Algorithm: Random Forest Regressor (200 estimators)

Target: ZIP-code food insecurity risk score (0–100)

Features:

Feature	Importance
Poverty Rate	68.7%
SNAP Participation Rate	18.5%
Food Desert Score	5.5%
Unemployment Rate	4.5%
Total Population	1.6%
Median Income	1.3%

Model Performance:

• Test MAE: ~0.72 points
• Cross-validated R²: 0.748 ± 0.083

---

🗄️ SQL Highlights

The sql/analytical_queries.sql file contains 22 production-quality queries, including:

• Window functions: LAG(), RANK(), NTILE(), AVG() OVER
• CTEs for multi-step transformations
• Multi-table JOINs across 4 data sources
• Aggregations and GROUP BY summaries
• Parameterized partner efficiency stored procedures

---

📺 Tableau Dashboard

Connect Tableau Public to tableau_exports/ CSVs to reproduce three dashboard views:

1. ZIP Risk Heatmap — Florida choropleth with Seminole County focus
2. Trend & Forecast — Time-series insecurity rates with COVID annotation
3. Partner Impact Leaderboard — Efficiency tier rankings

---

🌐 Live App Features

The Streamlit app provides four interactive pages:

• 🔍 ZIP Lookup — Enter any Central Florida ZIP → instant risk score, county comparison, factor breakdown, nearest partners
• 📊 County Dashboard — Ranked bar chart + poverty vs. insecurity scatter (67 counties)
• 🤝 Partner Analytics — Efficiency leaderboard by tier (Platinum → Bronze)
• 📈 Trends & Forecast — 5-county trend lines with 2-year linear forecast

---

🎯 Relevance to U.S. Hunger

JD Requirement	HungerSight Implementation
Tableau reports & dashboards	3-view Tableau Public dashboard
SQL queries, views, stored procedures	22 queries + 6 database views
Food insecurity statistics	67-county Florida dataset (2019–2023)
Partner data analytics	Composite efficiency scoring system
Actionable insights for stakeholders	Streamlit app + Tableau exports
Exploratory data research	ML feature importance analysis

---

👩‍💻 Author

Shivani Krishnama
Data Analytics Intern Portfolio Project
Central Florida · Built for U.S. Hunger, Longwood, FL

---

Data sources: U.S. Census Bureau · USDA ERS · Bureau of Labor Statistics · Feeding America