Machine Learning of CatBoost for Global Vertical Total Electron Content Prediction

Atmospheric molecules undergo partial ionization when exposed to sunlight, forming the ionosphere together with free electrons. The charged particles affect the passage of radio signals, leading to serious consequences such as signal loss in satellite navigation and disruptions in radio communication. In this study, we propose a window feature reconstruction algorithm based on the CatBoost model for global ionospheric prediction. We compare the feature-reconstructed CatBoost model with time series models, including the Long Short-Term Memory (LSTM) and Temporal Convolutional Network (TCN) models. The results demonstrate that the CatBoost model, enhanced by feature engineering, offers higher predictive accuracy for vertical total electron content (VTEC) forecasting.