Authors - Gokapay Dilip Kumar, Ravikumar Munaganuri, Dhruv Diguvapalli, Smayan Paul, A S Akhil, Mallu Shiva Rama Krishna, Prem Swarup Mallipudi, Kanchapogu Naga Raju, Sivanagaraju Vallabhuni Abstract - Solar flares are bursts that can disrupt terrestrial and space-based operations. This study aims to develop an unsupervised learning framework to anticipate flares and mitigate their impact. We applied clustering to unlabeled Solar Dynamics Observatory magnetic field time-series data to group similar activities and detect anomalies indicative of impending flares. Features were preprocessed and analyzed with K-Means and DBSCAN. Models differentiated regular activity from anomalous patterns preceding flares, revealing clusters with higher flare propensity. Unsupervised learning effectively forecasts solar flares by autonomously identifying critical patterns and yielded an MSE of 0.06. Integration of wind data and historical records with real-time prediction can be done in the future.
