Advancing Space Weather Prediction: Machine Learning and Bayesian Modelling for CMEs and Coronal Jets

This thesis investigates the use of Machine Learning (ML) and Bayesian inference to improve the prediction and understanding of Coronal Mass Ejection (CME), a critical aspect of space weather forecasting. Several ML techniques, including supervised learning methods such as support vector machines, decision trees, and ensemble methods, are used to develop predictive models based on CME data, aiming to enhance the accuracy of CME arrival time forecasts. A key focus is placed on model interpretability, achieved through Shapley Additive exPlanation (SHAP) values, which provide insights into the feature space and allow for a better understanding of how different variables influence model outputs. Additionally, the thesis applies Bayesian inference and Monte Carlo Markov Chain (MCMC) techniques to refine probabilistic models of CME propagation using drag-based models, further improving the robustness and reliability of the predictions. The work also extends ML applications to the study of other solar phenomena, specifically coronal jets, by augmenting the dataset for jet identification. This leads to increased dataset diversity, improved detection of rare events, and a better understanding of solar dynamics. Overall, this thesis presents advancements in the application of ML and Bayesian techniques to space weather forecasting and the study of solar phenomena. The tools and methods developed in this research hold considerable potential for future applications, with the capacity to improve prediction accuracy and mitigate the impacts of space weather on technological systems.