Conductivity prediction model for ionic liquids using machine learning

Area of Science:

• Materials Science
• Computational Chemistry
• Electrochemistry

Background:

• Ionic liquids (ILs) are versatile salts with applications in energy storage, solar cells, and sensors due to high ionic conductivity and thermal stability.
• Traditional physical methods for measuring IL conductivity are time-consuming and costly.
• Computational methods offer a rapid and effective alternative for screening IL properties.

Purpose of the Study:

• To develop a deep neural network (DNN) model for rapid and accurate prediction of ionic liquid conductivity.
• To create one of the most chemically diverse conductivity prediction models to date.
• To identify key chemo-structural features influencing ionic conductivity for designing new ILs.

Main Methods:

• Constructed a DNN model trained on 406 experimentally measured and published ionic liquid conductivity data points.
• Employed feature engineering to identify significant chemo-structural descriptors.
• Utilized a diverse dataset encompassing 406 unique and chemically varied ionic liquids.

Main Results:

• Achieved rapid and accurate predictions of ionic liquid conductivity using the developed DNN model.
• Identified key features correlating with ionic conductivity, serving as design guidelines for new ILs.
• Demonstrated improved performance over previous models constrained by data availability or IL diversity.

Conclusions:

• Machine learning, specifically DNNs, holds significant potential for accelerating the discovery and testing of high-conductivity ionic liquids.
• The developed model and identified features can guide the rational design of tailored ionic liquids for specific applications.
• This approach offers a cost-effective and efficient alternative to experimental measurements for IL conductivity screening.