Mimicking the halo-galaxy connection using machine learning

Paper highlights:

• Uses machine learning to predict galaxy baryonic properties from dark matter halo properties in the IllustrisTNG300 simulation.
• Combines multiple algorithms (tree-based methods, kNN, neural networks) and applies SMOGN data augmentation to address imbalanced datasets, improving predicted distributions and scatter.
• Achieves high accuracy for stellar mass and reasonable performance for other properties, reproducing galaxy power spectra for different galaxy populations.

You can read the paper here.