Stellar Flare Detection and Prediction Using Clustering and Machine Learning

A method for detecting and predicting stellar flares is presented using time-series data from NASA's TESS satellite. Unsupervised clustering with DBSCAN is first applied to the flux data to identify and label flare events as anomalies. Subsequently, an XGBoost classification model is trained on these labels to predict future flares, using lagged features to capture temporal dependencies. The XGBoost model shows promising results, successfully identifying smaller flares that a comparative LSTM model missed.