Volume 64

Machine Learning-driven Melting Point Prediction for Heat Recovery Materials Lizhong Yang, Xinyu You, Fei Xiao, Long Gao, Alessandro Romagnoli

Abstract

Facing the challenge of growing heating and cooling demand, more heat recovery materials used to capture and transfer heat should be developed. Melting point prediction is one of the most essential steps of machine-learning-driven material discovery. Existing methods rely on large datasets, often with the help of density function theory and molecular dynamics simulations, or complex deep learning methods. Here, we developed a multilayer perceptron model for predicting the melting point of the materials using a small but reliable dataset. By comparing with classical machine learning models, including ridge regression, random forest, XGBoost, and support vector machines, the multilayer perceptron model with transfer learning from a much larger dataset not dedicated to heat recovery materials demonstrated a flexible compromise balancing robustness and data size, providing a promising yet simple solution for melting point prediction.

Keywords heat recovery, melting point, machine learning, multilayer perceptron