This study focuses on the prediction of ionic conductivity in LLZO (Li₇La₃Zr₂O₁₂)-based solid electrolytes using machine learning techniques. A literature-based dataset was compiled, categorical variables were transformed into numerical representations, and a comprehensive data preprocessing workflow was applied.
During the modeling stage, a wide range of linear, nonlinear, and tree-based machine learning algorithms were evaluated. All models were optimized using GridSearchCV. While tree-based methods showed strong performance among individual models, the best overall results were obtained through ensemble learning. In this context, a weighted Voting Regressor model was constructed.
To improve the model’s ability to capture variable interactions, second-order feature interaction terms were also included. Following an error-based outlier analysis, the top 5% highest-error samples were removed, and the model achieved approximately R² ≈ 0.89 on the cleaned test dataset.
The findings indicate that ionic conductivity in LLZO systems is governed by nonlinear and complex interactions. This work was presented at the 3rd Bilsel International Erzurum Scientific Research Congress and prepared by Ozan ÖNAL
Congress Information:
3rd Bilsel International Erzurum Scientific Research Congress
25–26 April 2026, Erzurum/Türkiye
Authors:
Dr. Ozan ÖNAL
Publisher:
BILSEL PUBLISHING