Boosting-Crystal Graph Convolutional Neural Network for Predicting Highly Imbalanced Data: A Case Study for...

Area of Science:

• Materials Science
• Machine Learning
• Deep Learning

Background:

• Imbalanced datasets in materials science pose challenges for machine learning.
• Existing methods like oversampling can cause information loss or overfitting.

Purpose of the Study:

• To develop a deep learning framework for predicting minority-class materials, focusing on metal-insulator transition (MIT) materials.
• To address extreme class imbalances in materials data.

Main Methods:

• Introduced boosting-CGCNN, combining crystal graph convolutional neural network (CGCNN) with gradient boosting.
• Sequentially built a deeper neural network to handle class imbalances.

Main Results:

• The boosting-CGCNN model effectively handled extreme class imbalances in MIT material data.
• Demonstrated superior performance compared to existing approaches through comparative evaluations.

Conclusions:

• Boosting-CGCNN offers a promising solution for handling imbalanced datasets in materials science.
• The framework is particularly effective for predicting minority-class materials like MIT materials.