Prediction and Analysis of Ionic Conductivity in LLZO-Based Solid Electrolytes Using Machine Learning and Explainable Artificial Intelligence

Machine Learning and Explainable AI for Ionic Conductivity Prediction in LLZO-Based Solid Electrolytes
Project Overview

This interdisciplinary research project was developed by Dr. Ozan Önal as part of his doctoral research, integrating artificial intelligence, data science, and materials science.

The primary objective is to predict the ionic conductivity of LLZO (Li₇La₃Zr₂O₁₂)-based solid electrolytes using machine learning algorithms and to investigate the key factors influencing conductivity through explainable artificial intelligence techniques.

Research Motivation

Solid-state batteries represent a promising direction in advanced energy storage research. Solid electrolytes play an essential role in these technologies, and their ionic conductivity is a critical material property.

The ionic conductivity of LLZO-based electrolytes is influenced by several interacting factors, including chemical composition, dopant elements, stoichiometric ratios, synthesis conditions, and microstructural characteristics.

This research explores data-driven approaches to modeling these complex relationships and supporting the development of advanced solid electrolyte materials.

Dataset Development and Analysis

Experimental data reported in the scientific literature were collected and organized to develop a dataset of LLZO-based solid electrolytes.

The research workflow includes:

Collection and organization of experimental conductivity measurements.
Analysis of chemical compositions and dopant elements.
Data cleaning and consistency assessment.
Exploratory statistical analysis.
Feature engineering and preprocessing.
Investigation of synthetic data generation and parameter-based scenarios.
The dataset was used to investigate the relationships between material characteristics and ionic conductivity.

Machine Learning Methodology

Multiple regression algorithms were investigated and compared, including:

Linear Regression
Support Vector Regression (SVR)
Random Forest Regression
Gradient Boosting
XGBoost
Ensemble Learning
Model performance was evaluated using regression metrics such as R², Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE).

Comparative analysis was conducted to investigate the ability of different algorithms to capture linear and nonlinear relationships in the available data.

Explainable Artificial Intelligence

An important objective of the project is to improve the interpretability of machine learning predictions.

The SHAP (SHapley Additive exPlanations) framework was employed to examine the contributions of individual features to model predictions.

The explainability analysis focuses on:

Identifying influential model features.
Evaluating the contribution of chemical and experimental parameters.
Investigating positive and negative feature contributions.
Supporting the scientific interpretation of prediction results.
This approach improves model transparency and helps connect computational predictions with materials science knowledge.

Model Evaluation and Research Findings

Different modeling and data preparation strategies were evaluated during the research.

In selected experimental modeling scenarios, predictive performance reached an R² value of approximately 0.89.

These results should be interpreted in the context of the corresponding evaluation methodology, dataset characteristics, and validation strategy.

Computational scenario analyses were also investigated to evaluate possible combinations of material parameters and explore predicted conductivity behavior.

Scientific Contributions

The research contributes to the application of artificial intelligence in materials science through:

Data-driven prediction of ionic conductivity in LLZO-based solid electrolytes.
Comparative evaluation of machine learning regression algorithms.
Explainable AI analysis of influential material parameters.
Computational exploration of material composition scenarios.
Development of a research framework supporting solid-state electrolyte investigations.
Developer and Researcher

Dr. Ozan Önal

Artificial Intelligence Specialist | Data Scientist | Researcher

This project was developed by Dr. Ozan Önal within the scope of his doctoral research. Its main components include experimental data analysis, machine learning model development, comparative performance evaluation, and explainable artificial intelligence.

Research Disclaimer: This project is intended for scientific research. Machine learning predictions do not replace experimental measurements, and independent experimental validation is required for newly proposed material compositions.