Tree-Based Machine-Learning Classifier for Stellar Flares in The Zwicky Transient Facility Survey

This work is dedicated to solving the task of detecting flares from red dwarfs among the light curves of The Zwicky Transient Facility (ZTF) data. The study utilizes light curves with a temporal delay of no more than 30 min since the characteristic duration of flares ranges from 30 min to 2 h. The task is addressed using machine learning methods, specifically a binary classifier. Two models were employed as classifiers: random forest and gradient boosting. Both real ZTF data and synthesized flare light curves were used for model training. All models were tested on both synthesized flares and real flares which were found in the ZTF data previously. Based on the validation set with real flares, it was concluded that the gradient boosting model demonstrates the best performance. The achieved model quality allows it to be used for directly assembling a sample of red dwarf flare candidates. GitHub repository with full pipeline and demonstrative data is available at the link: https://github.com/snad-space/flare-classifier .