| Name: Debasish Mondal |
| Affiliation: Indian Institute of Science Education and Research (IISER) Tirupati |
| Conference ID: ASI2026_23 |
| Title: Machine learning based classification schemes for H I 21-cm absorbers |
| Abstract Type: Poster |
| Abstract Category: Galaxies and Cosmology |
| Author(s) and Co-Author(s) with Affiliation: Debasish Mondal(Indian Institute of Science Education and Research (IISER) Tirupati, Tirupati - 517619, India), Anirudh S. Nemmani(Nicolaus Copernicus Astronomical Center,Warsaw - PL-00-716, Poland), Arunima Banerjee(Indian Institute of Science Education and Research (IISER) Tirupati, Tirupati - 517619, India) |
| Abstract: H I 21-cm absorption traces cold atomic gas and may originate either in intervening galaxies or be associated with the background radio source. Since optical spectroscopy cannot feasibly classify the same for the large numbers of detections from blind surveys, we explore a machine-learning (ML) approach. Using 118 known H I 21-cm absorbers and spectral parameters from the Busy function fits, we train six ML models and find that a random forest provides the best performance (accuracy 89%, F1 = 0.9, AUC = 0.94). The linewidth parameter (w20) emerges as the most significant spectral feature. Moreover, a simplified random forest model using only w20 and the integrated optical depth performs nearly as well (accuracy 88%, F1 = 0.88, AUC = 0.91). Applying this model to 30 new absorbers from recent blind surveys (e.g. FLASH) demonstrates its utility for future large H I surveys with the Square Kilometre Array. |
