Data-Driven Optimization of Lithium NCM Cathode Material Using Machine-Learning Techniques

The optimisation of Lithium Nickel−Cobalt−Manganese (NCM) cathode materials is essential for improving the electrochemical performance of lithium-ion batteries. However, experimental evaluation of doped NCM compositions is slow, resource-intensive, and unable to efficiently explore the vast combinational space of dopants and synthesis parameters. This work presents a machine-learning-based predictive framework designed to estimate the discharge capacity of doped NCM cathode and cost-effectiveness. However, enhancing the discharge capacity and structural stability of NCM materials remains a complex scientific challenge. Their electrochemical behaviour is influenced by multiple interdependent factors, including dopant type, concentration, crystal structure, calcination conditions, and lattice distortion. Exploring such a multidimensional design space through traditional experimental methods is slow, expensive, and often yields limited insight due to the time-consuming nature of synthesis materials using material composition, structural characteristics, and synthesis-related data. Seven supervised regression algorithms—Linear Regression, Ridge Regression, Decision Tree Regression, Random Forest Regression, Gradient Boosting, Support Vector Regression, and K-Nearest Neighbor Regression—were compared to identify the most reliable model. Ridge Regression demonstrated the best balance of accuracy, stability, and generalisation, making it the final predictive model. The results show that machine learning can significantly reduce the dependency on costly laboratory experiments and accelerate the discovery of high- performance cathode compositions. This study contributes a data-driven methodology for supporting materials research and improving lithium-ion battery development.