Constructing a new hybrid model–Phy-XGBoost to predict the arrival time of coronal mass ejections

Predicting the arrival time of Coronal Mass Ejections (CMEs) is essential for space weather. In this study, we propose Phy-XGBoost, a physics-guided framework for CME arrival time prediction. We first extend the standard Drag-Based Model (DBM) to a Generalized Drag-Based Model (GDBM) to explore potential differences in the propagation of fast and slow CMEs, further refining it into a weighted GDBM (w-GDBM). To alleviate potential limitations of the physical model, we introduce XGBoost to predict the transit time residuals between observations and w-GDBM predictions using CME and solar-wind features. Across 800 random train test splits, Phy-XGBoost reduced the average MAE from 13.44 to 11.40 h and the average MedAE from 11.14 to 8.79 h compared with w-GDBM. Phy-XGBoost also achieved a comparable average MAE but a lower average MedAE than the purely data-driven XGBoost. These results suggest that the proposed physics-guided machine learning framework provides a promising approach to predicting CME arrival times.