Machine Learning for Cathode Materials in Secondary Metal-Ion Batteries
- 1Division for the Development of Systems for Informatics and ICT (ICT), Department for Energy Technologies and Renewable Sources (TERIN), ENEA—Italian National Agency for New Technologies, Energy and Sustainable Economic Development, C. R. Casaccia, Via Anguillarese, 301, 00123 Rome, Italy
- 2Division for the Development of Systems for Informatics and ICT (ICT), Department for Energy Technologies and Renewable Sources (TERIN), ENEA—Italian National Agency for New Technologies, Energy and Sustainable Economic Development, C. R. Bologna, Via dei Mille, 21, 40121 Bologna, Italy
Abstract
This review surveys machine learning approaches for cathode materials in secondary metal-ion batteries, covering classical algorithms, deep neural networks and graph neural networks, and linking them to the main cathode chemistries and electrochemical target properties. We provide a structured technical classification of model classes, their input representations, and the databases and descriptors most commonly used. We also discuss reproducibility and standardized benchmarking practices needed to support reliable comparisons between machine learning methodologies. Particular emphasis is placed on methodological factors affecting the physical consistency and transferability of graph neural network-based models for cathodes, including how transition metal chemistry affects the prediction of average voltage and formation energy. Across the reviewed approaches, dataset coverage, structural fidelity and the complexity of cathode chemistry emerge as major limitations to reliable property prediction. Finally, we discuss prospective directions, including the potential roles of large language models and quantum machine learning in data-driven cathode discovery.
Related
- Synergistic effect of Mn substitution and ball milling on NaCu 0.2 Fe 0.8− x Mn x O 2 cathode materials for sodium-ion batteries
- Investigation of NaxFePO4 systems prepared by a modified Pechini sol-gel method as possible cathode materials for high voltage Na-ion batteries
- Surfactant incorporation enhancing the two-electron transfer reaction of MnO2 cathode materials in aqueous Zn batteries
- Thermal Runaway Characteristics of Batteries with Manganese-Based Composite Cathode Materials Triggered by Electromagnetic Induction Heating and Lateral Heating
- Synthesis of Fused Biimidazole-Containing Polymers and Their Application as High-Potential Organic Cathode Materials for Lithium-Ion Batteries
BibTeX
@article{catillo2026machine,
title = {{Machine Learning for Cathode Materials in Secondary Metal-Ion Batteries}},
author = {Marco Catillo and Francesco Buonocore and Serena D’Onofrio and Simone Giusepponi and Massimo Celino},
journal = {Batteries},
year = {2026},
month = nov,
doi = {10.3390/batteries12110474},
url = {https://www.mdpi.com/2313-0105/12/11/474},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Machine Learning for Cathode Materials in Secondary Metal-Ion Batteries
AU - Marco Catillo
AU - Francesco Buonocore
AU - Serena D’Onofrio
AU - Simone Giusepponi
AU - Massimo Celino
PY - 2026
DA - 2026/11/06
JO - Batteries
DO - 10.3390/batteries12110474
UR - https://www.mdpi.com/2313-0105/12/11/474
AB - This review surveys machine learning approaches for cathode materials in secondary metal-ion batteries, covering classical algorithms, deep neural networks and graph neural networks, and linking them to the main cathode chemistries and electrochemical target properties. We provide a structured technical classification of model classes, their input representations, and the databases and descriptors most commonly used. We also discuss reproducibility and standardized benchmarking practices needed to support reliable comparisons between machine learning methodologies. Particular emphasis is placed on methodological factors affecting the physical consistency and transferability of graph neural network-based models for cathodes, including how transition metal chemistry affects the prediction of average voltage and formation energy. Across the reviewed approaches, dataset coverage, structural fidelity and the complexity of cathode chemistry emerge as major limitations to reliable property prediction. Finally, we discuss prospective directions, including the potential roles of large language models and quantum machine learning in data-driven cathode discovery.
SN - 2313-0105
ER -