High-entropy ceramic coatings have some unique physical and mechanical properties, such as high hardness, good corrosion resistance and excellent thermal stability. However, since they can contain five or more metal elements, their composition is quite complex. Combined with machine learning and high-throughput experimental methods, ultra-hard high-entropy ceramic coatings were screened in a short period of time. The hardness of coatings is predicted using a random forest algorithm based on its composition and processing parameters. The uncertainty of machine learning prediction is further reduced by active learning. After three iterations, a new high-entropy ceramic coating (AlCrNbTaTi)N with a hardness of 40.1 GPa has been successfully prepared, which is 9% higher than the optimal hardness of the original quinary system. This paper demonstrates that machine learning combined with high-throughput experimental methods can effectively accelerate design and composition optimization of multicomponent materials.