[Paper Review] Feature Analyses and Modelling of Lithium-ion Batteries Manufacturing based on Random Forest Classification

Kailong Liu, Xiao Hu+4 more|arXiv (Cornell University)|Feb 10, 2021

Advanced Battery Technologies ResearchEngineering36 references18 citations

TL;DR

This paper proposes a random forest (RF)-based framework to analyze and model lithium-ion battery manufacturing features, using out-of-bag predictions, Gini importance, and predictive measure of association (PMOA) to quantify feature importance and correlations. The method achieves high classification accuracy (macro F1 of 90.1% for electrode mass load and 66.4% for porosity), enabling interpretable sensitivity analysis and dimensionality reduction in battery manufacturing data.

ABSTRACT

Lithium-ion battery manufacturing is a highly complicated process with strongly coupled feature interdependencies, a feasible solution that can analyse feature variables within manufacturing chain and achieve reliable classification is thus urgently needed. This article proposes a random forest (RF)-based classification framework, through using the out of bag (OOB) predictions, Gini changes as well as predictive measure of association (PMOA), for effectively quantifying the importance and correlations of battery manufacturing features and their effects on the classification of electrode properties. Battery manufacturing data containing three intermediate product features from the mixing stage and one product parameter from the coating stage are analysed by the designed RF framework to investigate their effects on both the battery electrode active material mass load and porosity. Illustrative results demonstrate that the proposed RF framework not only achieves the reliable classification of electrode properties but also leads to the effective quantification of both manufacturing feature importance and correlations. This is the first time to design a systematic RF framework for simultaneously quantifying battery production feature importance and correlations by three various quantitative indicators including the unbiased feature importance (FI), gain improvement FI and PMOA, paving a promising solution to reduce model dimension and conduct efficient sensitivity analysis of battery manufacturing.

Motivation & Objective

• Address the lack of interpretable, data-driven methods for analyzing complex, interdependent features in lithium-ion battery manufacturing.
• Overcome limitations of trial-and-error approaches and black-box models in understanding how process variables affect electrode properties.
• Develop a systematic framework to simultaneously quantify feature importance and correlations for improved sensitivity analysis and model interpretability.
• Enable reliable classification of key electrode properties—mass load and porosity—using manufacturing data from mixing and coating stages.
• Provide a scalable solution for reducing model dimensionality and enhancing process understanding in battery production chains.

Proposed method

• Employ a random forest (RF) classifier trained on battery manufacturing data, leveraging out-of-bag (OOB) predictions for unbiased performance estimation.
• Use Gini importance to quantify the contribution of each feature (AMMC, StoLR, CG, viscosity) to classification accuracy of electrode mass load and porosity.
• Apply the predictive measure of association (PMOA) to estimate pairwise feature correlations, identifying strong dependencies between features.
• Integrate three distinct metrics—OOB error, Gini importance, and PMOA—into a unified framework for comprehensive feature analysis.
• Optimize hyperparameters of competing models (DT, KNN, SVM) using randomized search to ensure fair comparison with the proposed RF framework.
• Use 5-fold cross-validation to evaluate classification performance with macro F1 and AUC as primary metrics.

Experimental results

Research questions

• RQ1 How can feature importance and correlations be systematically quantified in lithium-ion battery manufacturing data using a data-driven approach?
• RQ2 To what extent do mixing-stage features (AMMC, StoLR, viscosity, CG) influence the classification of electrode mass load and porosity?
• RQ3 How does the proposed RF framework compare in classification performance to other machine learning models (e.g., DT, KNN, SVM) in this context?
• RQ4 What is the relative contribution of each manufacturing feature to the prediction of electrode mass load and porosity?
• RQ5 Can the framework support dimensionality reduction and sensitivity analysis by identifying key features and their interdependencies?

Key findings

• The proposed RF framework achieved a macro F1 score of 90.1% for classifying electrode mass load and 66.4% for porosity, outperforming DT, KNN, and SVM in both metrics.
• Coating weight (CG) was identified as the most important feature for electrode mass load, contributing to a 60.7% decrease in Gini impurity when removed.
• AMMC and StoLR showed high correlation (PMOA > 0.6) for both mass load and porosity, indicating strong functional relationships due to their shared role in slurry composition.
• Viscosity had low importance for porosity classification, suggesting it has limited direct impact on this property despite its role in slurry processing.
• The framework revealed that drying and calendering parameters are likely critical for porosity prediction, as current features alone yield suboptimal classification.
• The integration of OOB error, Gini importance, and PMOA provides a robust, interpretable method for feature analysis, enabling effective sensitivity and dimensionality reduction in battery manufacturing modeling.