Determining Importance of Features Related to Star Formation in Central and Satellite Galaxies Using Machine Learning
Key Definitions
Galaxy quenching: process in which star formation shuts down in a galaxy
Central galaxies: largest galaxy in center of halo
Satellites galaxies: smaller galaxies revolving around central galaxies
Spaxels: spatially resolved pixel containing vast information on a tiny area of the sky
Central galaxies: largest galaxy in center of halo
Satellites galaxies: smaller galaxies revolving around central galaxies
Spaxels: spatially resolved pixel containing vast information on a tiny area of the sky
Motivations
- Determine the most important parameter for predicting the quenching of galaxies on spatially resolved scales using ML algorithms
- Apply for subsets of the TNG100 galaxy sample including: high- and low- mass centrals, high- and low- mass satellites
- Compare results from TNG100 simulation data to observations from similar papers (see Bluck et. al 2020a)
Methods
- Conducting independent runs, utilizing cross validation resulting in 500 model runs
- Separation of training/testing on the galaxy-level to prevent data leakage
- Randomized Search CV utilized for hyperparameter tuning
- Utilized ROC curves, AUC/F1 scores to determine model accuracy
For Classification, our target variable was whether the spaxel quenched (0) or star-forming (1).
For Regression, our target variable was the star-formation rate of the spaxel.
For Regression, our target variable was the star-formation rate of the spaxel.