Abstract: Spike interference is one of the most common types of disturbance in 1-second geomagnetic sampling data, and its proper handling directly affects the quality of 1-second geomagnetic observations. Conventional methods, such as the median average approach require extensive manual intervention and exhibit limited adaptability when addressing interference lasting longer than one second. To achieve more accurate correction of spike interference, this study formulates the data correction task as a machine learning regression problem and proposes an XGBoost-based model for correcting spike disturbances in 1-second geomagnetic sampling data. Using raw observational data from the Z-component of the automated geomagnetic station system at the Manzhouli Seismic Observatory between January 2024 and April 2025, 17 720 samples of magnetic field variations—during and around spike interference events—were extracted to construct the dataset for model training. The model achieved the following performance metrics on the test set: Mean Squared Error (MSE) of 0.011 667 nT², Root Mean Squared Error (RMSE) of 0.104 742 nT, Mean Absolute Error (MAE) of 0.076 582 nT, and Coefficient of Determination (R²) of 0.999 852. The results demonstrate the excellent performance of the XGBoost model in correcting spike interference in 1-second geomagnetic sampling data, providing an effective new method for automated and high-precision correction of such disturbances.