Evaluating Machine Learning Methods

Publication

Our research has been accepted for presentation at the 16th Congress of the International Colour Association (AIC 2025), taking place October 19-24, 2025, in Taipei, Taiwan. The conference theme "Color for the Future" aligns perfectly with our investigation into modern machine learning approaches for color prediction in printing processes.

Introduction

This project investigates the effectiveness of various machine learning models in predicting XYZ color values from CMY input data, compared to traditional polynomial regression methods. Utilizing datasets such as FOGRA51, APTEC(PC10), and APTEC(PC11), the research explores the accuracy of machine learning techniques including Gradient Boosting and Random Forest.

Objectives

• To bridge the gap in current literature by introducing a comparative analysis of modern printer datasets.
• To systematically compare a suite of machine learning algorithms to polynomial regression in color characterization tasks.

Methodology

• Data Collection: Utilized datasets from the International Color Consortium.
• Machine Learning Methods: Bayesian Methods, Decision Trees, Deep Learning, Elastic Net, Gradient Boosting, k-Nearest Neighbors, Lasso Regression, Random Forest Regressor, Ridge Regression, Simple Multilayer Perceptron (MLP), and Support Vector Machine (SVM).
• Polynomial Regression: Implemented for degrees ranging from 1 to 10.
• Evaluation: Error metrics calculated using the CIEDE2000 formula.

Results

Machine learning models demonstrated varying degrees of improvement in error metrics over polynomial regression. However, the improvements were relatively modest, suggesting that the benefits of machine learning must be carefully balanced against their complexity.

Conclusion

The study highlights the potential of machine learning in enhancing color prediction accuracy in printing processes. Future research could further explore the transition from 3-channel LAB to 4-channel CMYK data for more comprehensive color mapping.

Tools and Technologies Used

• PyCharm
• Python
• Color Science Library
• Scikit-learn (sklearn)
• Additional Libraries: NumPy, Pandas, Matplotlib

Challenges and Solutions

Faced challenges such as limited dataset size and model selection, addressed by experimenting with various settings and models.

How to Run

• Clone the repository.
• Install the required dependencies: pip install -r requirements.txt
• Run the main script: python Main.py

License

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

Acknowledgments

Thanks to all the datasets providers and the open-source community for making this research possible.