Classification of the magnetic state is an essential step to investigate two-dimensional magnetic materials. Combining high-throughput calculations and machine-learning methods, we have classified the magnetic states of 23 825 MXenes in the aNANt database. A simple descriptor, obtained by averaging the product of the element feature, connectivity, and Coulomb matrix, was found to improve the performance of the machine-learning models. Using this descriptor on 4153 data produced using first-principles calculations, predictive machine-learning models were developed and 1432 MXene with a high saturation magnetization were predicted. The proposed descriptor is useful for the magnetic classification of other materials, and the identified magnetic MXene materials can be used as an important reference for further study.