Abstract Details
| Name: Abhiram K Affiliation: Indian Institute of Technology Dharwad Conference ID: ASI2025_620 Title : Automatic Classification and Anomaly Detection of Supernova Spectra in ZTF Bright Transient Survey Authors and Co-Authors : Abhiram Krishna 1, Yujing Qin 2, Christoffer Fremling 2, Shrinivas R Kulkarni 2 Abstract Type : Poster Abstract Category : High Energy Phenomena, Fundamental Physics and Astronomy Abstract : The Zwicky Transient Facility (ZTF) Bright Transient Survey (BTS) represents a significant effort in identifying and characterizing extragalactic transients through a comprehensive and unbiased spectroscopic approach. One of the primary missions is to acquire optical spectra to classify extragalactic transients, including subclasses of Supernovae and Tidal Disruption Events (TDEs). We develop an automatic supernovae classification tool using machine learning by designing a series of binary classifiers to classify different types of supernovae using ensemble classifiers, primarily the Random Forest Classifier and XGBoost Classifier. A hierarchical tree classifier is also designed by chaining various binary classifiers to achieve multi-class classification of different supernova subtypes. Furthermore, the extensive collection of spectra from the survey and the community would allow the detection of rare and new classes of transients that can deepen our understanding of the evolution of stars and activities of supermassive black holes. We employ an unsupervised Isolation Forest algorithm and various dimensionality reduction techniques to identify these novel, unusual events through spectroscopy. This approach allows for the detection of significant anomalies, potentially revealing rare and extraordinary supernovae within the dataset. Future surveys, such as 4MOST, DESI-II, and MSE, promise to expand the transient dataset significantly, making manual inspection impractical. Automated, robust classification pipelines, such as those developed in this work, will be essential for managing and fully utilizing these unprecedented datasets. |