Machine Learning Study on the Electron Emission Properties of the Thermionic Cathode Impregnants

The emission current density of thermionic cathode materials is predicted using machine learning methods. Firstly, four classical regression algorithms of linear regression (LR), random forest regression (RFR), XGBoost, and gradient boosting regression tree (GBRT) are used. The relation coefficient
$\boldsymbol{R}^{\mathbf{2}}$
is chosen for the model evaluation metrics, and the prediction results are 0.174, 0.677, 0.812, and 0.915, respectively. Secondly, to solve the problems of small training data and unstable prediction results, the small sample TrAdaboost regression algorithm is introduced to predict and adjust the parameters for each system with different percentages of each chemical component (612, 512, 411, and others), and the final prediction results
$\mathbf{R}^{\mathbf{2}}$
are obtained as 0.954, 0.95, 0.5 and 0.939. The prediction accuracy is significantly improved.