A machine learning approach to estimating HI deficiency in galaxies

Measurements of the HI content of galaxies serve as an important tracer for probing the impact of environment on galaxy evolution. More specifically, the HI deficiency (defined as the difference between expected unaltered and observed HI content of a galaxy) is closely related with environmental effects, which are most significant in large groups and clusters. In this work, we aim to estimate the HI deficiency of ALFALFA galaxies and investigate its relation with galactic environment. Using a random forest machine learning algorithm, we developed a predictive model capable of estimating the original HI content of a galaxy based solely on its optical properties. The model was trained on a subsample of 6 982 isolated ALFALFA galaxies with optical photometric data from the Sloan Digital Sky Survey (SDSS). Our predictive model outperforms the traditional approach, in which HI mass is linearly related to optical size (both on a logarithmic scale). The model achieves RMSE $\approx$ 0.22 dex and $R^2 \approx$ 0.80, compared with RMSE $\approx$ 0.26 dex and $R^2 \approx$ 0.70 for the traditional method. We applied this model to predict the expected HI content for non-isolated ALFALFA galaxies, enabling the calculation of HI deficiency. Controlling for the effects of internal factors, like stellar mass and presence of AGN, we find an increase in binned median HI deficiency of 0.15 dex attributable to environmental effects. In addition, we evaluate the temporal evolution of the predicted HI mass, and associated HI deficiency, due to the evolving stellar populations, following a gas removal event.