Predicting Halo Formation Time Using Machine Learning
Context:Halo formation time, which quantifies the mass assembly history of dark-matter halos, directly impacts galaxy properties and evolution. Although not directly observable, it can be inferred through proxies like star formation history or galaxy spatial distributions. Recent advances in machine learning enable more accurate predictions of halo formation time using galaxy and halo properties. Aims:This study aims to investigate a machine learning-based approach to predict halo formation time-defined as the epoch when a halo accretes half of its current mass-using both halo and baryonic properties derived from cosmological simulations. By incorporating properties associated with the brightest cluster galaxy located at the cluster center, its associated intracluster light component and satellite galaxies, we aim to surpass these analytical predictions, improve prediction accuracy and identify key properties that can provide the best proxy for the halo assembly history. Methods:Using The Three Hundred cosmological simulations, we train Random Forest (RF) and Convolutional Neural Network (CNN) models on halo and baryonic properties, such as mass, concentration, stellar and gas masses, and features of the brightest cluster galaxy and intracluster light. CNN models are trained on two-dimensional radial property maps. We also construct simple linear models using only observationally accessible features. Results:RF models show median biases of 4%-9% with standard deviations of 20%. CNN models reduce median bias to <4%, although they have higher scatter. Simple linear models using a limited number of observables achieve prediction accuracy comparable to RF models. Traditional relations between halo formation time and mass/concentration are preserved.