Feature Blending: An Approach toward Generalized Machine Learning Models for Property Prediction.

Area of Science:

• Materials Science
• Computational Chemistry
• Machine Learning

Background:

• Machine learning (ML) is increasingly used in materials science for tasks like property prediction, but a generalized framework is lacking.
• Existing ML models for property prediction are often highly accurate but lack broad applicability across different material classes.

Purpose of the Study:

• To develop a generalized machine learning framework for predicting material properties across diverse material classes.
• To introduce a novel feature selection approach enabling a unified ML model for property prediction.

Main Methods:

• Proposed a feature selection approach involving blending selected features from various data classes to create global data descriptors.
• Developed a unified ML model using Gaussian process regression (GPR) with the blended features.
• Utilized accurate band gaps of three distinct classes of 2D materials as the target property for model development.

Main Results:

• The developed ML model achieved an average root-mean-squared error of 0.12 eV for predicting band gaps of unseen 2D materials.
• The feature blending approach successfully captured both class-specific and global material traits.

Conclusions:

• The proposed feature blending approach enables the development of a unified ML model for accurate property prediction across different material classes.
• This methodology can be extended to incorporate additional material classes and predict other material properties.