An integrated machine learning model for accurate and robust prediction of superconducting critical temperature

Discovering new superconductors via traditional trial-and-error experimental approaches is apparently a time-consuming process, and the correlations between the critical temperature (T) and material features are still obscure. The rise of machine learning (ML) technology provides new opportunities to speed up inefficient exploration processes, and could potentially uncover new hints on the unclear correlations. In this work, we utilize open-source materials data, ML models, and data mining methods to explore the correlation between the chemical features and T values of superconducting materials. To further improve the prediction accuracy, a new model is created by integrating three basic algorithms, showing an enhanced accuracy with the coefficient of determination (R) score of 95.9 % and root mean square error (RMSE) of 6.3 K. The average marginal contributions of material features towards T values are estimated to determine the importance of various features during prediction processes. The results suggest that the range thermal conductivity plays a critical role in T prediction among all element features. Furthermore, the integrated ML model is utilized to screen out potential twenty superconducting materials with T values beyond 50.0 K. This study provides insights towards T prediction to accelerate the exploration of potential high-T superconductors.