Comparative Analysis of Machine learning techniques for Forecasting Ionospheric Total Electron Content Data

Abstract

The ionosphere is a highly dynamic region of the Earth's atmosphere that plays a crucial role in global navigation and communication systems. Accurate forecasting of ionospheric activity is essential for mitigating its impact on these systems. In recent years, machine learning techniques have shown promise in predicting ionospheric activity, but there is limited research on their comparative performance. This paper presents a comparative analysis of various machine learning techniques for forecasting ionospheric total electron content (TEC) data. Specifically, we compare the performance of five popular machine learning techniques- linear regression, multi-layer perceptron neural networks, K-nearest neighbors, support vector regression and random forest regressor. We use TEC data along with exogenous parameters namely By, Bz, Vp, Np, F10.7, Kp, Dst and Ap. We evaluate the performance of the models at different latitudes and during solar quiet and active years. Our results show that the Random Forest Regressor (RFR) outperformed the other techniques with the lowest root mean square error (RMSE) and mean absolute error (MAE). The R2 value suggests that the RFR model provides the best fit to the TEC data compared to other models evaluated and can be used for ionospheric TEC forecasting.