Identifying single spectral lines by combining different machine learning strategies

The future Euclid and WFIRST surveys will be covering a large portion of the extragalactic sky in near-IR slitless spectroscopy. A relevant fraction of the spectra collected will show only one emission line in the wavelength range covered. In order to maximize the scientific return of these missions, it is important that single emission lines are identified correctly. To this purpose, we developed and combined supervised and un-supervised machine learning algorithms. Our software is calibrated and tested on a ``gold'' sample selection of WISP data characerized by secure identifications. WISPs (WFC3 IR Spectroscopic Parallel survey) represents one of the most important proxies of the future Euclid and WFIRST surveys. We run the algorithm on a sample of WISP single line sources, identifying an originally missed class of sources characterized by small size and high [OIII]/Hα flux ratios flux ratios