Machine-learning model for prediction of martensitic transformation temperature in NiMnSn-based ferromagnetic shape memory alloys

• XGBRegressor (XGBR) model is demonstrated to accurately predict the martensitic transformation temperature ( T M ) of NiMnSn-based ferromagnetic shape memory alloys (FSMAs). • We find that a combination of features Numa , Arc , and avg Ven can effectively improve the fitting effect ( R 2 = 0.903) of XGBR model. • K-fold Cross-Validation is used to prove that the XGBR model shows high generalization ability ( R 5 f 2 = 0.869 and R 3 f 2 = 0.838) on small data sets and can be used to predict unknown T M of NiMnSn-based FSMAs. Martensitic transformation temperature ( T M ) of NiMnSn-based ferromagnetic shape memory alloys (FSMAs) is crucial to identifying the operating range of an application. From a materials design point of view, an efficient method that can predict the T M accurately should be strongly pursued, to meet various applications with different operating temperatures. In this paper, we demonstrate that machine learning (ML) can rapidly and accurately predict the T M in NiMnSn-based FSMAs. We evaluate the performance of four machine learning models, including Random Forest Regressor (RFR), Support Vector Regression (SVR), Linear Regression (LR), and XGBRegressor (XGBR) model. Three important features of Numa , Arc , and avg Ven are selected as the optimal feature combination for building the model. Moreover, to ensure the best generalization ability of the model, multiple methods of cross-validation (Leave-One-Out Cross-Validation, 3-fold Cross-Validation, and 5-fold Cross-Validation) are used. Finally, the XGBR model exhibits the best performance for predicting the T M ( R 2 = 0.903 and RMSE = 5.4, R 5 f 2 = 0.869 and R 3 f 2 = 0.838). The results of small deviation and variance proven that the XGBR model, proposed in this work, is suitable to be used to predict the T M of unknown NiMnSn-based FSMAs. This work is expected to promote the targeted design of FSMAs.