Predicting viscosity of ionic liquids using structure-based models.

Area of Science:

• Materials Science
• Computational Chemistry
• Chemical Engineering

Background:

• Ionic liquids (ILs) exhibit tunable structures, leading to a broad spectrum of viscosities.
• Predicting IL viscosity is crucial for their application but challenging due to structural diversity.

Purpose of the Study:

• To develop and compare machine learning models for predicting the dynamic viscosity of ionic liquids.
• To establish accurate structure-property relationships for ILs.

Main Methods:

• Utilized cheminformatics (RDKit) to generate molecular descriptors for ILs.
• Developed two structure-based models: an Artificial Neural Network (ANN) with KMeans clustering and dimensionality reduction, and a Gradient Boosting method (CATBoost) using all descriptors.
• Employed the elbow methodology for hyperparameter optimization in the ANN model.

Main Results:

• Both ANN and CATBoost models demonstrated strong generalizability across various ILs and viscosity ranges.
• The ANN model achieved R² values of 0.94 (training) and 0.87 (test).
• The CATBoost model achieved R² values of 0.94 (training) and 0.85 (test), indicating robust prediction capabilities for both.

Conclusions:

• Machine learning, particularly ANN with dimensionality reduction and CATBoost, effectively predicts ionic liquid viscosity.
• The developed models establish reliable structure-property relationships, applicable to designing ILs with specific viscosities.
• This approach can be extended to predict other properties of soft materials based on their molecular structures.