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.

Model Details

Developed by: ChanMeng666
Model type: Multi-layer perceptron (MLP) neural network
Language(s): Python
License: Apache-2.0
Finetuned from model: No - trained from scratch

Model Description

This modelcard aims to be a base template for body fat percentage estimation using anthropometric measurements. Two variants are provided:

Full Input Model

14 input features
20 neurons in hidden layer
Achieves R² = 0.9724, MSE = 1.9250
Optimal for maximum accuracy

Reduced Input Model

9 input features
5 neurons in hidden layer
Achieves R² = 0.9617, MSE = 2.6734
Optimized for efficiency

Both models use sigmoid activation for hidden layer and linear activation for output layer.

Uses

Direct Use

The models can be used directly for body fat percentage estimation from anthropometric measurements:

# Example usage
import tensorflow as keras

model = keras.models.load_model('best_full_model.keras')
predictions = model.predict(measurements)

Downstream Use

Integration into health and fitness applications
Clinical body composition assessment
Research studies on body composition
Personal health monitoring systems

Training Details

Training Data

Dataset contains anthropometric measurements including:
- Body circumference measurements (neck, chest, abdomen, hip, thigh, knee, etc.)
- Height and weight
- Body density measurements
Data split: 60% training, 20% validation, 20% test

Training Procedure

Optimizer: Adam with learning rate 0.1
Loss function: Mean Squared Error
Batch size: 32
Early stopping with patience 100
Random seed: 123 for reproducibility

Evaluation

Testing Data, Factors & Metrics

Evaluated on 20% held-out test set
Metrics: R² score and Mean Squared Error (MSE)
Results validated across different body types and measurement ranges

Model Performance

Model Type	R² (All)	MSE (All)	R² (Test)	MSE (Test)
Full Input	0.9724	1.9250	0.9671	1.5301
Reduced Input	0.9617	2.6734	0.9394	2.8187

Bias, Risks, and Limitations

Models require accurate anthropometric measurements
Performance may vary across different demographic groups
Not intended to replace medical assessments
Results should be interpreted by qualified professionals
Model accuracy depends on measurement precision

Recommendations

Users (both direct and downstream) should:

Ensure accurate measurement techniques
Consider using the full input model when maximum accuracy is required
Use the reduced input model when efficiency is prioritized
Validate results against other assessment methods
Consider demographic and individual factors in interpretation

Technical Specifications

Model Architecture

Input layer: 14 features (full) or 9 features (reduced)
Hidden layer: 20 neurons (full) or 5 neurons (reduced) with sigmoid activation
Output layer: 1 neuron with linear activation

Hardware Used

Training compatible with standard CPU
GPU optional for faster training
Minimal memory requirements

Software Requirements

TensorFlow 2.0+
Python 3.7+
Keras 2.0+
NumPy
Pandas

Citation

If you use this model, please cite it as:

@software{ChanMeng666_2024_bodyfat_estimation_mlp,
  author = {ChanMeng666},
  title = {Body Fat Estimation using Neural Networks},
  year = {2024},
  publisher = {Hugging Face},
  url = {https://huggingface.co/ChanMeng666/bodyfat-estimation-mlp}
}