🧠 Medical Machine Learning: Predicting and Correlating Diabetes Rates in San Diego

🚀 Goal
Analyze and model the prevalence of diabetes across San Diego County using Big Data and Machine Learning, focusing on potential racial correlations. We’ve built an analytical model, performed demographic correlation analysis, and created an interactive website to visualize our insights.

🔗 Website: Visit Project Website

---

📌 Project Summary

This project investigates how race correlates with the prevalence of diabetes across 17 cities in San Diego County. Our approach combines geospatial visualization, statistical modeling, and machine learning—specifically an overdetermined least squares regression optimized via gradient descent. The final model outputs Race Linked Coefficients (RLCs) which are used to estimate diabetes prevalence in other populations.

---

📊 Tools & Technologies

• 🗺️ Kepler.gl – Geospatial visualization
• 📊 Tableau – Interactive charts & dashboards
• 🧠 Python – Data wrangling, regression modeling, neural network simulation
• 🧮 MATLAB – Matrix computations, backpropagation
• 🧑‍💻 VSCode – Code development
• 🌐 Google Sites – Interactive infographic website

---

🧬 Target Demographics

• Hispanic/Latino
• Black
• Asian Pacific Islander (API)
• White

Population samples include only these four racial groups.

---

🏙️ Target Cities

17 cities in San Diego County were selected based on data availability, including:

• Lemon Grove
• Chula Vista
• National City
• ... (full list available on website)

---

⚕️ Defining a Diabetes Case

A diabetes "case" is any event involving:

• Death
• ER Admission
• Hospitalization
• In-patient Treatment
• Physical Rehabilitation
• SNF/Nursing Home

---

🔍 Key Findings

• Top 3 Cities with Highest Diabetes Prevalence:

  1. Lemon Grove
  2. Chula Vista
  3. National City

• These cities had Hispanic/Latino as the dominant demographic group (>50%).

• Correlations Observed:

  • 🧑🏿‍🦱 Black: Strong positive correlation
  • 🧑🏽 Hispanic/Latino: Moderate positive correlation
  • 🧑🏻 White: Moderate negative correlation
  • 🧑🏼‍🦱 API: Moderate positive correlation

---

🧮 Machine Learning Model

🔗 Race Linked Coefficients (RLCs)

A unique scalar value per race to model:

🧠 Neural Network Inspired Optimization

• Solves an overdetermined system using least squares regression
• Optimized via gradient descent with backpropagation
• Model achieves ~95% accuracy on training data
• Enables future forecasting based on population growth trends

---

⚠️ Data & Assumptions

• Adults aged 18+
• Population sample limited to 4 racial groups
• Confounding variable: Age distribution not controlled (age 65+ more vulnerable)
• Cases per 100,000 = (Total Cases ÷ Population) × 100,000

---

📚 Research Inspiration

Inspired by work from the Technical University of Munich on neural networks solving linear systems.

---

👥 Meet the Team

Name	LinkedIn	GitHub
Siddhant Borse	LinkedIn	GitHub
Pranjal Patel	LinkedIn	GitHub
Jesse Hurtado	LinkedIn	GitHub
Atharva Kulkarni	LinkedIn	GitHub

---

📌 Future Improvements

• Incorporate age and income as confounding variables
• Add forecast automation using demographic projection APIs
• Explore classification models for high-risk population segmentation

---

📄 License

MIT License © 2025 — Medical ML Team

---

⭐ Show Your Support!

Give this repo a ⭐ if you found it insightful or useful. Feel free to fork and build upon it!