Zirconia ceramics are unique in their transformation-induced plasticity (TRIP), allowing plastic strains of up to 10%, along with unique shape memory and superelastic properties. However, the current lack of interatomic potentials for doped ZrO₂ ceramics, which can accurately capture phase transitions and deformation mechanisms, limits our atomic-level comprehension of the TRIP effect in these ceramics. In our research, we introduce a transfer learning-based deep neural network interatomic potential (DP-Zr_1-xCe_xO₂) that attains an accuracy on par with hybrid metaGGA DFT calculations. This potential successfully predicts phase transitions in both pure ZrO₂ and ZrO₂-CeO₂ ceramics with quantitative agreement with experiments. Leveraging this potential, we conduct large-scale MD simulations to explore the plasticity, shape memory and superelastic characteristics of ZrO₂ ceramics under complex loading conditions.