README.md

metadata

title: BodyFatPredictor
emoji: 🧍‍♂️
colorFrom: red
colorTo: red
sdk: docker
app_port: 8501
tags:
  - streamlit
pinned: false
short_description: Streamlit app with BodyFat (%) predictor
license: mit

🧍‍♂️ Body Density Predictor – Ridge Regression

This project presents a machine learning regression application that predicts human body density from anthropometric measurements using a Ridge Regression model.
The implementation follows the original ML Olympiad – Predicting Wellness competition setup, where body density is the primary target variable.

---

📌 Project Overview

The main objectives of this project are:

• Predict body density (Density) from physical body measurements
• Build a stable and interpretable regression model for a small dataset
• Provide an interactive Streamlit demo for single and batch predictions
• Clearly communicate model assumptions and limitations

This project is intended for educational and portfolio purposes.

---

📊 Dataset & Target

• Input features:
  Anthropometric measurements such as age, height, weight, and body circumferences (neck, chest, abdomen, hip, thigh, etc.)

• Target variable:
  Density (human body density)

The model is trained directly on Density, without predicting body fat percentage.

---

🧠 Model Description

• Model type: Ridge Regression (L2-regularized linear regression)
• Reason for model choice:
  • Handles strong multicollinearity between body measurements
  • Performs well on small datasets
  • Produces stable and smooth predictions
  • Easy to interpret and deploy

---

🧮 Feature Engineering

Only features that were used during training are applied in the app:

• Waist-to-Hip Ratio:
  Waist_hip = Abdomen / Hip

markdown Code kopieren

• Body Index (BMI, imperial units):
  Body_Index = 703 × Weight / Height²

yaml Code kopieren

Engineered features are calculated only if required by the trained model.

---

📏 Units & Input Conventions

Measurement	Unit
Height	Centimeters (cm) (converted internally to inches)
Weight	Pounds (lbs)
Body circumferences	Inches
Model output	Density (unitless)

⚠️ Correct units are essential for meaningful predictions.

---

🖥️ Application Features

🔹 Single Prediction

• Interactive form with realistic input ranges
• Automatic unit conversion (cm → inches)
• Feature engineering and alignment
• Density prediction with plausibility warnings
• Display of final model inputs

🔹 Batch Prediction (CSV)

• Upload CSV files with required feature columns
• Automatic feature alignment
• Download predictions as a CSV file

---

⚠️ Important Notes on Predictions

• Human body density typically lies within a very narrow range (≈ 0.95 – 1.10)
• Predictions outside this range may indicate:
• Unrealistic input values
• Inputs outside the training distribution

The app provides warnings when predicted density values fall outside the typical physiological range.

---

🧪 Limitations

• Small dataset size limits generalization
• Linear regression may oversimplify complex physiological relationships
• Predictions may be unreliable for extreme or unrealistic input values
• The model is not intended for medical or clinical use

These limitations are explicitly communicated in the app interface.

---

🚫 Disclaimer

This application is not a medical device.
All predictions are provided for educational and demonstration purposes only and must not be used for health diagnosis or treatment decisions.

---

🧰 Tech Stack

• Python
• pandas, NumPy
• scikit-learn
• Streamlit
• joblib

---

📎 Repository Structure

├── app.py ├── ridge_model.pkl ├── requirements.txt └── README.md

yaml Code kopieren

---

✅ Conclusion

This project demonstrates a complete and transparent machine learning pipeline for predicting body density using Ridge Regression.
By focusing directly on the competition target variable and avoiding unnecessary post-processing, the application provides stable, interpretable, and scientifically consistent predictions, making it well suited as a portfolio and learning project.

---