We investigate the problem of estimating quantum field statistics, specifically the mean photon number (n¯) and the zero time-delay second-order coherence function (g(2)(0)), for an unknown single-mode field. The field is assumed to interact with a two-level atom via the Jaynes–Cummings Hamiltonian, and the resulting population inversion dynamics are analyzed using three tree-based machine learning models, such as the Random Forests, Extra Trees, and Gradient Boosting. We find that these models predict n¯ and g(2)(0) of the unknown initial field state with high accuracies, as quantified by the coefficient of determination (R2) and the root-mean-square error (RMSE). Our framework of implementing machine learning models for characterizing quantum states in a simple model of atom–field interaction can be a step toward automated state characterization in complex quantum systems.