Machine Learning-Driven Prediction of Thermal Conductivity in Fe Alloys

Thermal conductivity is an essential feature that affects boilers and heat transfer devices, among other applications to transport heat effectively in fe alloys. The conductivity of heat has generally been determined by laborious and costly experimental techniques. In consideration of material composition data, the current work studies the application of machine learning (ML) models to estimate thermal conductivity in Fe alloys. We evaluate the efficacy of many regression models, such as Support Vector Machines (Regression), Linear Regressor, Random Forest, XGBoost, etc., The findings show that amongst all other used machine learning algorithms CatBoost is capable of accurately and successfully predicting thermal conductivity. The CatBoost model’s average squared error (MSE), average absolute error (MAE), and average absolute percentage error (MAPE) were 3.44, 1.37, and 0.049 respectively on the test set with coefficient of determination $\mathbf{R 2}$ value of 0.97. This study lays the path for using ML to speed up the production of novel Fe alloys with customized thermal conductivity characteristics.