The exploration of fast fluoride-ion conductors, which can be applied as solid electrolytes of all-solid-state fluoride-ion batteries, requires generalized material design principles. Herein, we used regression learning based on compositional descriptors to predict fluoride-ion conductivity, revealing that random forest regression, mainly considering cation polarizability, enables conductivity prediction in unexplored compositional spaces. Composition-based material exploration aided in identifying compounds with high fluoride-ion conductivity, as exemplified by a tysonite-type species (Ba 0.2 Sn 0.8 F 2 ; 9.0 × 10 −5 S cm −1 at 298 K), and substitution with La resulted in a further increase in conductivity to 1.4 × 10 −4 S cm −1 at 298 K (Ba 0.175 Sn 0.775 La 0.05 F 2.05 ). Thus, we established design principles and accelerated the exploration of fast fluoride-ion conductors using compositional information.