Paper Number
ICIS2025-2191
Paper Type
Complete
Abstract
AI-based battery health prediction is a key technology for sustainable energy management, yet systematic evaluation of model performance and model–feature interaction effects remain limited. We conducted a comprehensive meta-analysis from 199 experimental records across 35 studies. Our findings indicate that in scenarios where high predictive accuracy is needed, hybrid deep learning models are the preferred choice, whereas for scenarios requiring lightweight deployment or better interpretability, random forest emerges as the more suitable option. Moreover, model performance is highly dependent on feature compatibility. We validated the robustness of results through sensitivity analysis and publication bias test and proposed a five-step guideline to help practitioners dynamically align business needs, model architectures, and feature combinations. This study provides empirical guidance for efficient model development in battery health prediction and enhances data-driven decision-making in sustainable energy management.
Recommended Citation
Wang, Yifan; Kost, Leonard; Heumann, Maximilian; and Schade, Wolfgang, "Navigating Model Selection and Feature Engineering for Battery Health Prediction: An Evidence-Based Meta-Analysis" (2025). ICIS 2025 Proceedings. 10.
https://aisel.aisnet.org/icis2025/da_bus/da_bus/10
Navigating Model Selection and Feature Engineering for Battery Health Prediction: An Evidence-Based Meta-Analysis
AI-based battery health prediction is a key technology for sustainable energy management, yet systematic evaluation of model performance and model–feature interaction effects remain limited. We conducted a comprehensive meta-analysis from 199 experimental records across 35 studies. Our findings indicate that in scenarios where high predictive accuracy is needed, hybrid deep learning models are the preferred choice, whereas for scenarios requiring lightweight deployment or better interpretability, random forest emerges as the more suitable option. Moreover, model performance is highly dependent on feature compatibility. We validated the robustness of results through sensitivity analysis and publication bias test and proposed a five-step guideline to help practitioners dynamically align business needs, model architectures, and feature combinations. This study provides empirical guidance for efficient model development in battery health prediction and enhances data-driven decision-making in sustainable energy management.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
07-DataAnalytics