Comparing HistGradientBoost vs. LightGBM for Kp Geomagnetic Index Prediction

The Sun's solar wind and interplanetary magnetic field interact continuously with the Earth's magnetosphere, which causes geomagnetic disturbances that are characterized using the Kp index. Accurate Kp index predictions are thus crucial for mitigating the risks space weather poses to global technological infrastructure. In this work, a comparative study of two state-of-the-art machine learning algorithms, HistGradientBoost and LightGBM, for forecasting the Kp index is presented. Using several decades of data (1963-2024) from the Low-Resolution OMNI (LRO) dataset¹, our optimized LightGBM model achieved a predictive R² of 0.984. A detailed comparison of the models will be presented, along with a feature-importance analysis which identifies the most critical predictors.