CGMacros: Machine Learning & Visualization for Personalized Nutrition

Uncover meaningful relationships between dietary habits, glucose responses, physical activity, and gut health.

TABLEAU Link for Visual Insights: https://public.tableau.com/app/profile/venkata.botta/viz/02DataExplorers_Visuals_17514099877070/DataExplorersStoryboard

Project Overview

This is a comprehensive data science project analyzing the CGMacros dataset, a rich, multimodal dataset of 45 adults tracked over 10 days with continuous glucose monitoring, dietary macronutrients, physical activity, and gut microbiome profiles. The objective is to uncover meaningful relationships across four interconnected health domains to enable personalized nutrition strategies and early identification of metabolic risk.

Objectives

• Conduct exploratory data analysis across diet, glucose, activity, and gut microbiome domains
• Build machine learning models to predict postprandial glucose responses
• Identify actionable patterns and correlations for personalized health interventions
• Present findings through interactive visualizations and insights

Key Findings

Demographic Profile

• 45 participants tracked over 10 days
• 64% female, 51% obese, 75% Hispanic/Latino
• Majority in 50s age group
• Diverse baseline risk factors

Dietary Impact

• Breakfast causes the steepest glucose spikes (200+ mg/dL)
• Macronutrient ratios matter significantly:
  • Higher protein, fat, and fiber relative to carbs reduce glucose spikes
  • Glycemic response moderated by nutrient composition

Physical Activity Benefits

• Post-meal activity consistently blunts glucose spikes across all groups
• Most dramatic benefits observed in:
  • Diabetic participants
  • Obese participants
• Powerful, non-invasive tool for glycemic regulation

Gut Health Integration

• Better microbial balance correlates with:
  • Lower fasting glucose
  • Lower insulin levels
• Gut microbiome acts as an upstream driver of metabolic outcomes
• Essential for long-term metabolic control

Machine Learning Results

• Models tested: Linear Regression, KNN, Random Forest, XGBoost, LightGBM
• Best performer: LightGBM
  • R² Score: 0.78
  • RMSE: 16.99 mg/dL
• Successfully predicts 2-hour postprandial glucose responses

Key Discoveries

1. Metabolic health is multifactorial—no single factor determines glucose control
2. Holistic screening matters—even normal-weight individuals with optimal cholesterol can exhibit dysglycemia
3. Personalized macronutrient strategies significantly impact glucose outcomes
4. Post-meal activity is a practical, evidence-based intervention
5. Gut microbiome composition plays a vital role in metabolic health