Missing value imputation for > 2 MeV electron fluxes in geostationary orbit based on GA-RF model

A genetic algorithm-optimized random forest algorithm (GA-RF) model is constructed to impute large-scale missing data for 5-min averaged data of > 2 MeV electron integral fluxes from GOES-E/W satellites. The model inputs include V, Vx, the proton density, the SYM/H index, B, B_x, B_y, B_z, AU, AE, and > 0.6 MeV and > 2 MeV electron integral fluxes from GOES-E/W. The target variable is the > 2 MeV electron integral flux from GOES-W/E. A comparison of the GA-RF model with other machine learning models, including the backpropagation (BP), long short-term memory (LSTM), random forest (RF), extreme learning machine (ELM), and extreme gradient boosting (XGBoost) models, reveals that the GA-RF model achieves the highest PE and LC values and the lowest RMSE and MAE values, indicating that the GA-RF model outperforms the other models in imputing large-scale missing data. Compared with commonly used interpolation methods, such as cubic spline interpolation and linear interpolation, the GA-RF model effectively captures electron flux variations and provides imputed data that closely align with satellite-detected values.