Artificial Intelligence and Machine Learning

Systematic Feature Design for Cycle Life Prediction of Lithium-ion Batteries During Formation Rhyu et al demonstrate how to filter high-dimensional raw current-voltage-time electrochemical data for regions of high statistical significance, produce candidate features and then down select based on domain knowledge, and then finally prune down to data-driven statistical features that can be directly explained by mechanistic models to ensure robustness and potentially allow extrapolation to new materials with minimal data.

Using only formation data (from Cui et al), cycle life during accelerated aging can be predicted within 10% error using just 2 mechanistically informed features in a random-forest model, directly connecting statistical modeling with theory to enable explainable and trustable predictions. Source: Rhyu et al, Joule, 2025, 95, 101884

Extracted from the PDF; charts and images may read imperfectly.