Body Fat Prediction using Neural Networks

This project implements advanced neural network models for accurate prediction of body fat percentage using anthropometric measurements. Through comprehensive analysis and optimization, we've developed both full-feature and reduced-input models that achieve high accuracy while maintaining practical applicability.

Features

🧠 Advanced Neural Architecture

• Multi-layer perceptron models with optimized hidden layer configurations
• Support for both comprehensive and reduced-input feature sets
• Adaptive learning with early stopping and optimization

📊 Comprehensive Analysis Suite

• Detailed correlation analysis of body measurements
• Feature importance evaluation through sensitivity analysis
• Extensive model performance comparisons

🎯 High Prediction Accuracy

• R² score of 0.9724 for full-feature model
• MSE as low as 1.9250 on test data
• Robust performance across different body types

💡 Smart Feature Selection

• Intelligent reduction of input measurements
• Maintains high accuracy with fewer required measurements
• Practical implementation considerations

📈 Performance Visualization

• Detailed performance metrics and comparisons
• Feature correlation heatmaps
• Model sensitivity analysis plots

Getting Started

1. Clone the repository:

git clone https://github.com/ChanMeng666/bodyfat-estimation-mlp.git
cd bodyfat-estimation-mlp

2. Install required packages:

pip install -r requirements.txt

3. Run the Jupyter notebooks:

jupyter notebook

Model Performance

Model Type	R² Score	MSE	Hidden Layers
Full Input	0.9724	1.9250	20
Reduced Input	0.9617	2.6734	5

Tech Stack

Project Structure

├── notebooks/
│   ├── Part3_(i)_Qualitative_Analysis.ipynb
│   ├── Part3_(ii)_Network_Performance.ipynb
│   ├── Part3_(iii)_Correlation_Analysis.ipynb
│   ├── Part3_(iv)_Reduced_Input_Model.ipynb
│   ├── Part3_(v)_Sensitivity_Analysis.ipynb
│   ├── Part3_(vi)_Performance_Comparison.ipynb
│   └── Part3_(vii)_Summary.ipynb
├── data/
│   └── Body_Fat.csv
├── models/
│   ├── best_full_model.keras
│   └── best_reduced_model.keras
├── requirements.txt
└── README.md

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Dataset sourced from clinical body composition measurements
• Research methodology based on neural network optimization techniques
• Performance metrics and analysis methods from established machine learning practices

For detailed implementation and analysis, please refer to the individual notebooks in the repository.