SAIP2026

Predicting Crystal Symmetry in Sodium-Ion Battery Materials Using Ensemble Machine Learning Models

7 Jul 2026, 12:20

20m

Lecture Hall GH1 (University of the Western Cape)

Oral Presentation Track A - Physics of Condensed Matter and Materials Physics of Condensed Matter and Materials

Speaker

Keletso Monareng (University of Limpopo)

Description

Keletso Monareng1, Petros Ntoahae1, and Rapela Maphanga2,3
1Department of Physics, University of Limpopo, Private Bag X 1106, Sovenga, 0727, Polokwane, South Africa
2Renewable and Sustainable Energy Research Centre, Sol Plaatje University, Private Bag X 5008, Kimberly, 8300, South Africa
3 National Institute for Theoretical and Computational Sciences, NITheCS, Gauteng, 2000, South Africa

Abstract

Sodium-ion batteries (SIBs) have emerged as a promising alternative to lithium-ion technologies due to the natural abundance, low cost, and wide geographical availability of sodium. These advantages make SIBs particularly attractive for large-scale energy storage applications. However, the discovery and optimisation of suitable electrode materials remains challenging due to the vast compositional and crystallographic design space and the complex relationships between composition, structure, and electrochemical performance. Traditional experimental and density functional theory approaches are computationally expensive and time-consuming. In this study, a machine learning-driven framework is developed using engineered compositional features, including stoichiometric coefficients, ionization energy, oxidation states, and ionic radii. Feature importance analysis reveals that compositional coefficients and ionization energy contribute most significantly to predictive performance, while oxidation states and ionic radii provide complementary contributions. Model performance was evaluated across multiple algorithms, including Random Forest, K-Nearest Neighbours, Decision Tree, Gradient Boosting, Extreme Gradient Boost, Light Gradient Boosting Machine, and Multi Linear Perceptron, for predicting crystallographic properties such as crystal system, Bravais lattice, point group, and space group. Ensemble boosting models outperform others, with Extreme Gradient Boost and Light Gradient Boosting machine achieving the highest performance, reaching Weighted Balanced Accuracy values of approximately 0.95-0.96 and Weighted Matthews Correlation Coefficient values above 0.90. Overall, the Light Gradient Boosting Machine emerges as the best-performing model, enabling rapid and reliable screening of candidate materials and accelerating the discovery of SIB electrode materials.