🌱 Sustainability of Machine Learning Models

This repository provides an implementation of a genetic programming algorithm to optimize a multi-objective function, ensuring the sustainability of machine learning (ML) models. The rationale behind balancing accuracy, interpretability, and energy efficiency is to serve as example promoting responsible AI development.

🚀 Overview

Genetic Programming (GP) is an evolutionary algorithm inspired by biological evolution. It searches for mathematical expressions that optimize predefined objectives. This project applies GP to evolve ML models with a focus on three key aspects:

1. Prediction Accuracy – Measured by Mean Squared Error (MSE) on a test dataset.
2. Model Simplicity – Encouraging interpretable solutions.
3. Energy Efficiency – Minimizing computational power consumption.

This approach offers a practical reference for sustainable computing and can be adapted to various AI applications.

📥 Installation

Ensure you have Python installed, then install the required dependencies using:

pip install numpy deap scikit-learn

🔧 Usage

1. Clone the repository:

   git clone /~https://github.com/jorge-martinez-gil/sustainability.git
   cd sustainability

2. Run the script:

   python sustainability.py

The script will:

• Load a sample dataset.
• Split it into training and testing sets.
• Execute the genetic programming algorithm.
• Output the best evolved models with their respective performance scores (MSE, simplicity, and energy efficiency).

This implementation provides a foundation for further improvements, such as refining the interpretability metric or improving energy consumption calculations.

📚 Research & Related Work

This project is based on research into sustainable semantic similarity models, integrating accuracy, interpretability, and energy efficiency.

For more details, refer to:

@article{martinez2022sustainable,
  title={Sustainable semantic similarity assessment},
  author={Martinez-Gil, Jorge and Chaves-Gonzalez, Jose Manuel},
  journal={Journal of Intelligent & Fuzzy Systems},
  volume={43},
  number={5},
  pages={6163--6174},
  year={2022},
  publisher={IOS Press}
}

📄 License

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