Predicting the Geomagnetic Activity Indicator Kp via Gradient-Boosted Decision Trees: A Comparative Assessment of HistGradientBoost and LightGBM

Predicting the geomagnetic index Kp, a scale that measures the intensity of a geomagnetic storm,is essential for understanding, mitigating, and lessening space-weather impacts on Earth’s technologicalinfrastructure. Recent advances in machine learning has provided powerful tools to improveprediction accuracy. In this paper we examine the effectiveness two modern gradient-boosting algorithms- HistGradientBoost (Histogram-based Gradient Boosting) and LightGBM - for forecastingthe Kp index over an extended data set spanning several solar cycles (1963 - 2024) via the NASAOMNI dataset [7]. Both methods outperform a persistence baseline, with LightGBM achievingslightly higher accuracy: MAE= 0.833, MSE= 1.26, and R2 = 0.984 on the held-out test set (Ourdata had a 80 20 test split). We discuss the strengths and limitations of each approach and outlinefuture work involving additional data sources and hybrid models.