Attention-Enhanced Graph Neural Networks for Accurate Prediction of Elastic Properties in Crystalline Materials

DOI

10.2991/978-94-6463-922-3_7How to use a DOI?

Keywords

Elastic Property Prediction; Graph Neural Network (GNN); Additive Attention Mechanism; SHAP Analysis; Elastic Anisotropy

Abstract

Accurately predicting the elastic characteristics of crystalline materials from compositional and structural data is a significant issue in materials informatics. Conventional machine learning methods, such as Linear Regression, Random Forest, and XGBoost, often fail to capture the complex and non-linear interactions seen in crystalline systems, leading in low predicted accuracy. To address these constraints, we offer a new Graph Neural Network with Additive Attention (GNN With Attention) that builds a homogenous graph using k-nearest neighbour (k-NN) feature similarity and uses an attention strategy to prioritise relevant nodes during pooling. This strategy allows the model to learn more detailed, structure-aware representations of materials. We compare the model to five baselines—Linear Regression (LR), Random Forest (RF), XGBoost (XGB), Fully Connected Neural Network (FFNN), and a vanilla GNN—for five target elastic properties: bulk modulus (K_VRH), shear modulus (G_VRH), energy per atom, Poisson's ratio, and elastic anisotropy. The suggested model outperforms baseline models like Linear Regression (R² = 0.7916), Random Forest (R² = 0.1493), and XGBoost (R² = 0.7500), achieving better results with RMSE = 22.8789, MAE = 10.4728, and R² = 0.8778. The SHAP study verifies the model's capacity to capture complicated feature relationships. To summarise, the GNN With Attention framework provides a robust and interpretable solution for predicting elastic characteristics in crystalline materials, opening the path for rapid material discovery.

Copyright

© 2025 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - P. Lakshmi Rajan
AU  - Sujatha Nair
AU  - M. S. Arun
PY  - 2025
DA  - 2025/12/25
TI  - Attention-Enhanced Graph Neural Networks for Accurate Prediction of Elastic Properties in Crystalline Materials
BT  - Proceedings of the International Conference Recent Advances in Materials, Processes and Technology for Sustainability (RAMPTS 2025)
PB  - Atlantis Press
SP  - 93
EP  - 124
SN  - 2590-3217
UR  - https://doi.org/10.2991/978-94-6463-922-3_7
DO  - 10.2991/978-94-6463-922-3_7
ID  - Rajan2025
ER  -