Scalable, Advanced Machine Learning-based Approaches for Stellar Flare Identification: Application to TESS short-cadence Data and Analysis of a New Flare Catalogue

We apply multi-algorithm machine learning models to TESS 2-minute survey data from Sectors 1-72 to identify stellar flares. Models trained with Deep Neural Network, Random Forest, and XGBoost algorithms, respectively, utilized four flare light curve characteristics as input features. Model performance is evaluated using accuracy, precision, recall, and F1-score metrics, all exceeding 94%. Validation against previously reported TESS M dwarf flare identifications showed that our models successfully recovered over 92% of the flares while detecting $\sim2,000$ more small events, thus extending the detection sensitivity of previous work. After processing 1.3 million light curves, our models identified nearly 18,000 flare stars and 250,000 flares. We present an extensive catalog documenting both flare and stellar properties. We find strong correlations in total flare energy and flare amplitude with color, in agreement with previous studies. Flare frequency distributions are analyzed, refining power-law slopes for flare behavior with the frequency uncertainties due to the detection incompleteness of low-amplitude events. We determine rotation periods for $\sim120,000$ stars thus yielding the relationship between rotation period and flare activity. We find that the transition in rotation period between the saturated and unsaturated regimes in flare energy coincides with the same transition in rotation period separating the saturated and unsaturated levels in coronal X-ray emission. We find that X-ray emission increases more rapidly with flare luminosity in earlier-type and unsaturated stars, indicating more efficient coronal heating in these objects. Additionally, we detect flares in white dwarfs and hot subdwarfs that are likely arising from unresolved low-mass companions.