Work

Halo Coronal Mass Ejection Recognition Using Aditya-L1 Data and Machine Learning

Year: 2026
Type: conference-paper
Abstract: This work moves forward a new machine learning strategy broadly featuring multiple learning algorithms that can operate with data about space weather phenomena. In order to alleviate noisy and incomplete measurements, we have transformed CACTUS event logs and NASA CDF datasets into our resources. The event logs brought out the features that mattered most and proton density and bulk speed were among the features. Our ensemble methods, mainly Isolation Forest and Random Forest classifiers, were employed to locate abnormal CME events, and, to determine their effects, by discovering outliers and measuring their severity. The classification stage was instrumental in cleaning the datasets for easier visualization, whereby both rule-based and quantile-based severity labels were used. The experimental results demonstrated an overall classification accuracy of 75%, thus, the method had a strong ability in finding event windows and making severity predictions with confidence. The findings here serve as evidence of the feasibility of hybrid machine learning solutions in the prompt identification of space weather events. Additional investigation will employ deep learning models to achieve more precise event detection and to offer more adaptable satellite data modeling. Such a methodology will undoubtedly be central to the development of more effective automated forecasts of solar-induced disturbances.
Language: English
Open Access status: hybrid
OpenAlex ID: W7202201177
FWCI:
Cites: 3
Cited by:
Related to: 10