Lithium-Ion Battery Property Prediction Using GNN and Active Learning

Project Overview

This project focuses on predicting lithium-ion battery properties using advanced machine learning techniques, including Graph Neural Networks (GNNs), SHAP analysis, and active learning strategies. The pipeline is structured to provide robust predictions, interpret model behavior, and enhance performance using minimal data via active learning.

Project Contents

• src/: Source code for the project, including data processing, GNN modeling, SHAP analysis, and active learning.
  • data_preparation.py: Handles loading, preprocessing, and saving of data.
  • model_definition.py: Defines the GNN model architecture and graph representation.
  • train.py: Implements the training loop for the GNN and active learning strategies.
  • evaluate.py: Evaluates the trained model and generates visualizations.
  • shap_analysis.py: Conducts SHAP-based feature importance analysis.
• notebooks/: Google Colab notebooks are provided in the notebook folder for direct and accurate implementation of these models.

Cite this work:

If you use this repository, codes in your research or work, please cite the following publication:

Das, D. and Chakraborty, D. (2025). Machine Learning Prediction of Physicochemical Properties in Lithium-Ion Battery Electrolytes With Active Learning Applied to Graph Neural Networks. Journal of Computational Chemistry, 46: e70009.
https://doi.org/10.1002/jcc.70009

Features

• Graph Neural Network (GNN) Modeling:

  • Predicts key properties:
    • Electronic Energy
    • Total Enthalpy
    • Total Entropy
    • Free Energy
    • Vibration Frequencies
  • molecular graph representations as input.

• Active Learning Framework:

  • Implements strategies to select the most informative data points:
    • Maximum Expected Improvement (MEI)
    • Maximum Uncertainty (MU)

• SHAP Analysis:

  • Provides interpretability by explaining feature contributions to model predictions.

• Validation Dataset:

  • Evaluates model performance on independent datasets for better generalization.

• Visualization:

  • Plots for actual vs. predicted values.
  • Training and validation learning curves.
  • SHAP-based feature importance visualizations.

Processed Data (Utilized for this study)

This project uses the Lithium-Ion Battery (LIB) Electrolyte datasets, containing detailed properties of lithium-ion battery electrolytes. These are the final, processed datasets, adapted from the original datasets for training and evaluating the models developed in this project.

Dataset Link: LIBE Dataset

Tools and Technologies

• Python 3.8+: Core programming language.
• PyTorch & PyTorch Geometric: For GNN modeling.
• SHAP: For feature importance analysis and interpretability.
• Matplotlib & Seaborn: For data visualization.
• Scikit-learn: For metrics and preprocessing.
• Pandas & NumPy: For data manipulation and numerical operations.
• Google Colab: For Computation on GPU

Original Datasets

The project utilizes the Lithium-Ion Battery Electrolyte (LIBE) dataset and MPcules dataset, which contains comprehensive data related to the properties and performance of lithium-ion battery electrolytes.

Citation:

1. Spotte-Smith, Evan Walter Clark; Blau, Samuel M.; Xie, Xiaowei; Patel, Hetal; Wood, Brandon; Dwaraknath, Shyam; et al. (2021). Lithium-Ion Battery Electrolyte (LIBE) dataset. figshare. Dataset. https://doi.org/10.6084/m9.figshare.14226464.v2
2. Spotte-Smith, E. W. C.; Cohen, O. A.; Blau, S. M.; Munro, J. M.; Yang, R.; Guha, R. D.; Patel, H. D.; Vijay, S.; Huck, P.; Kingsbury, R.; et al. A database of molecular properties integrated in the Materials Project. Digital Discovery 2023, 2 (6), 1862-1882,

Tools and Technologies

• Python 3.8: Core programming language used for the project.
• NumPy: For numerical computations.
• Pandas: For data manipulation and analysis.
• Matplotlib & Seaborn: For data visualization.
• scikit-learn: For machine learning algorithms and data preprocessing.
• PyTorch & torch-geometric: For building and training Graph Neural Networks.
• pymatgen: For materials science data manipulation.
• SHAP: For explainable AI and model interpretability.

Dependencies and Installation

Before running the code, install the following Python libraries:

pip install torch pip install shap==0.46.0 pip install scikit-learn==1.5.2 pip install matplotlib pip install seaborn pip install pandas pip install numpy pip install pymatgen pip install matminer==0.9.3 pip install torch-geometric==2.6.1

Project Structure

LIBE/
├── README.md                  # Documentation (this file)
├── notebooks/
│   ├── GNN_Training.ipynb     # Jupyter notebook for step-by-step training
├── src/
│   ├── data_preparation.py    # Data preprocessing and graph creation
│   ├── model_definition.py    # GNN model definition
│   ├── train.py               # Training and active learning
│   ├── evaluate.py            # Evaluation metrics and result visualization
│   ├── shap_analysis.py       # SHAP feature importance analysis
├── results/
│   ├── models/                # Trained models saved here
│   ├── plots/                 # Generated plots saved here
├── LICENSE                    # BSD-2-Clause License
└── .gitignore                 # Git ignore file