Ternary pnictide semiconductors with II−IV−V 2 stoichiometry hold potential as cost-effective thermoelectric materials with suitable electronic transport properties, but their lattice thermal conductivities (κ) are typically too high. Insights into their vibrational properties are therefore crucial to finding strategies to reduce κ and achieve improved thermoelectric performance. We present a theoretical exploration of the lattice thermal conductivities for a set of pnictide semiconductors with ABX 2 composition (A = Zn, Cd; B = Si, Ge, Sn; and X = P, As) using machine-learning-based regression algorithms to extract force constants from a reduced number of density functional theory simulations and then solving the Boltzmann transport equation for phonons. Our results align well with available experimental data, decreasing the mean absolute error by ∼3 W m −1 K −1 with respect to the best previous set of theoretical predictions. Zn-based ternary pnictides have, on average, more than double the thermal conductivity of the Cd-based compounds. Anisotropic behavior increases with the mass difference between A and B cations, but while the nature of the anion does not affect the structural anisotropy, the thermal conductivity anisotropy is typically higher for arsenides than for phosphides. We identify compounds such as CdGeAs 2 , for which nanostructuring to an affordable range of particle sizes could lead to κ values low enough for thermoelectric applications.