Star formation and morphological properties of galaxies in the Pan-STARRS $3 \pi$ survey- I. A machine learning approach to galaxy and supernova classification

We present a classification of galaxies in the Pan-STARRS1 (PS1) 3$\pi$ survey based on their recent star formation history and morphology. Specifically, we train and test two Random Forest (RF) classifiers using photometric features (colors and moments) from the PS1 data release 2. Labels for the morphological classification are taken from Huertas-Company+2011, while labels for the star formation fraction (SFF) are from the Blanton+2005 catalog. We find that colors provide more predictive accuracy than photometric moments. We morphologically classify galaxies as either early- or late-type, and our RF model achieves a 78\% classification accuracy. Our second model classifies galaxies as having either a low-to-moderate or high SFF. This model achieves an 89\% classification accuracy. We apply both RF classifiers to the entire PS1 $3\pi$ dataset, allowing us to assign two scores to each PS1 source: $P_\mathrm{HSFF}$, which quantifies the probability of having a high SFF, and $P_\mathrm{spiral}$, which quantifies the probability of having a late-type morphology. Finally, as a proof of concept, we apply our classification framework to supernova (SN) host-galaxies from the Zwicky Transient Factory and the Lick Observatory Supernova Search samples. We show that by selecting on $P_\mathrm{HSFF}$ or $P_\mathrm{spiral}$ it is possible to significantly enhance or suppress the fraction of core-collapse SNe (or thermonuclear SNe) in the sample with respect to random guessing. This result demonstrates how contextual information can aid transient classifications at the time of first detection. In the current era of spectroscopically-starved time-domain astronomy, prompt automated classification is paramount.