Title: Using Machine Learning to Isolate and Fit Optical Emission Lines of Gaseous Nebulae

The growing volume of archived spectral data together with the anticipated volumes of data from future telescope missions necessitate the development of new, efficient methods of analyzing spectral data. Machine learning technologies present one possible approach to automatic and effective spectra analysis. In this work, we demonstrate a machine learning–based approach to automatically identifying and characterizing features of interest in optical emission spectroscopy data of gaseous nebulae. To start, the continuum must be identified, fit and removed from the spectral data. After analyzing the continuum, a trained classifier model differentiates between the spectral emission lines and noise in the signal. With the emission lines identified, a clustering algorithm is used to isolate each peak before it is fit to a Voigt profile. From the optimized profile, the characteristic values (width, height, central wavelength, etc.) are determined for each peak identified within the spectra. The characteristics of each line can be matched to a corresponding atomic transition that emits light at an optical wavelength. This approach is demonstrated by analyzing archived HST/STIS spectra of the binary star Eta Carinae.