Predicting the heat capacity of strontium-praseodymium oxysilicate SrPr4(SiO4)3O using machine learning, deep learning, and hybrid models

Abstract

• ML/DL models accurately predict the specific heat capacity of SrPr 4 (SiO 4 ) 3 O. • A Hybrid model combining RF and DBN models outperforms other ones in accuracy. • The application of machine/deep learning in thermophysical property prediction. • The Hybrid model gives an RMSE of 0.47 and R 2 of 0.9999 in heat capacity prediction. • The Leverage method validated the integrity of the data used and hybrid model A thorough examination of the specific heat capacity of strontium-praseodymium oxysilicate, which serves as a vital thermophysical parameter was investigated. This parameter is important in enhancing the performance and efficiency of heat transfer-based equipment, as well as applications in catalysis, insulation materials, and advanced ceramics. Machine learning (ML) offers a potent solution for forecasting diverse processes using both data-driven and knowledge-based methods. In this study, the capability of five advanced machine learning models, including Random Forest (RF), Gradient Boosting (GBoost), Extreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), and Decision Tree (DT) models, and three deep learning models, TabNet, Deep Belief Network (DBN), and Deep Neural Network (DNN) was investigated. Our analysis indicates that the Random Forest and Deep Belief Network models outperform all other competing models. Additionally, we introduce a hybrid model combining these two models, which enhances the accuracy of predicting the heat capacity of strontium-praseodymium oxysilicate. Specifically, the Hybrid model achieved an AAPRE of 0.42213, an RMSE of 0.38914, and a near-perfect R 2 value of 0.9999. The analysis indicated that the Hybrid model could accurately anticipate how the heat capacity of strontium-praseodymium oxysilicate would change. In conclusion, outlier detection was performed using the Leverage method to identify any anomalous data points, thereby illustrating the effective range of the proposed Hybrid model.

Topics & Concepts