This project leverages the Body Fat Prediction dataset from Kaggle, which includes anthropometric measurements (e.g., height, weight, waist) to predict body fat percentage using machine learning.

🎯 Build a machine learning model to estimate body fat percentage from physical measurements.

• 🏋️ Fitness & Health: Personalize diet and exercise plans.
• 🏥 Medical Diagnosis: Assess risks for obesity-related conditions.
• 🏆 Sports Optimization: Ideal body composition for athletic performance.
• 💰 Insurance: Risk-based premium adjustment.

• Motivation: Replace expensive tools (DEXA) with scalable solutions.
• Use Case: Gyms or fitness apps provide instant assessments.
• Value: Low-cost health insights boost engagement and subscriptions.

• Use Case: Scalable B2B solutions for clinics, schools, or corporations.
• Value: Enables health monitoring at scale without medical devices.

• Use Case: Tailored plans (meals, workouts, apparel) based on body fat.
• Value: Higher customer satisfaction → more conversions and loyalty.

• Use Case: More accurate risk profiling than BMI alone.
• Value: Better risk management for insurers, better outcomes for health programs.

• 📂 Source: Kaggle – Body Fat Prediction
• 💾 Size: 436 samples, 16 columns
• 🧪 Features:
  • Target: BodyFat (percentage)
  • Inputs: Age, Weight, Height, Abdomen, Chest, Neck, Thigh, Hip, etc.

• Bodyfat EDA Methodology

• Dataset was clean, numeric, and no nulls.

• Visualizations revealed normal distributions with minor skewness.

• Strong correlations:

  • Negative: BodyFat vs Density
  • Positive: Abdomen & Chest vs BodyFat

• Sex-based analysis showed anatomical differences, but Sex had a weak impact on body fat prediction.

📈 Key Insights:

• Abdomen circumference is the strongest single predictor.
• Weight correlates more with muscle mass than fat.

Custom features:

• bmi = Weight / (Height/100)^2
• waist_to_hip = Abdomen / Hip
• waist_to_height = Abdomen / Height
• arm_ratio = Forearm / Biceps

Started simple due to data size:

• ✅ Linear Regression (baseline)
• ✅ SVR + RFE + PCA (enhanced model)
• ✅ Separate models: Male / Female / Combined

Metrics:

• MAE, RMSE, R²
• Evidently AI reports for:
  • Data Drift
  • Target Drift
  • Regression Performance

• ⚙️ Data Processing: Polars, Scikit-Learn
• 🔁 Workflow: Metaflow
• 📊 Experiment Tracking: MLflow
• 🔍 Monitoring: Evidently AI
• 🧪 Tuning: Optuna
• 📦 API: FastAPI
• 📤 Serving: Streamlit dashboard
• 📂 Model Storage: joblib + MLflow Artifacts

# Docker Deployment
docker-compose up --build   # First time
docker-compose up           # Regular run

# Local Access
http://localhost:8000       # Web form
http://localhost:8000/docs  # Swagger UI

Sample curl request:
curl -X POST "http://localhost:8000/predict/" \
     -H "accept: application/json" \
     -H "Content-Type: application/x-www-form-urlencoded" \
     -d "abdomen=110&hip=120&weight=100&thigh=190&knee=50&biceps=38&neck=45"

• ✅ Accurate prediction of body fat percentage
• ✅ Feature-based health insights using anthropometric measurements
• ✅ Scalable FastAPI for real-world integration (web & REST)
• ✅ Visual analytics using Evidently AI for drift detection and retraining triggers