Synergistic Modeling of Liquid Properties: Integrating Neural Network-Derived Molecular Features with Modified Kernel...

Area of Science:

• Computational Chemistry
• Machine Learning
• Physical Chemistry

Background:

• Scarcity of experimental data challenges machine learning (ML) applications in computational chemistry.
• Increasing complexity of ML models exacerbates the need for data-efficient methods.

Purpose of the Study:

• To develop a robust ML approach for predicting liquid properties despite limited experimental data.
• To integrate theoretical principles into ML models for enhanced accuracy and interpretability.

Main Methods:

• Deriving molecular features from a complex neural network (NN) model.
• Applying these features to a simpler, conventional ML model robust to overfitting.
• Incorporating Arrhenius temperature dependence into a modified kernel model.

Main Results:

• Successfully predicted liquid properties such as viscosity and surface tension.
• The modified kernel model showed significant improvements in specific scenarios.
• Demonstrated the utility of NN-derived features for conventional ML models.

Conclusions:

• The proposed method effectively addresses the challenge of data scarcity in computational chemistry ML.
• The integration of theoretical concepts enhances the performance and applicability of ML models.
• The approach offers a promising direction for predicting molecular system properties.