The emission peak wavelengths of Eu2+-doped inorganic phosphors have been predicted from only the stoichiometric information of the host materials. The prediction models were constructed with machine-learning technology optimized for screening inorganic phosphors. They are sufficiently simple and practical to explore the host materials. The error in the predicted emission wavelengths of the phosphors is 139 meV over the entire visible region. The models were theoretically validated by relating them to the current understanding of the mechanism controlling emission spectra.