Abstract: Machine learning techniques were introduced into the preparation and characterization of cermet powders to predict their performance. Raw material particle size, binder content, atomization frequency and sintering temperature were taken as input features, while spherical particle size and powder flowability of the product were set as output targets. Gaussian process regression and Bayesian ridge regression models were constructed. After data augmentation, hyperparameter optimization via leave-one-out cross-validation and model training with five-fold cross-validation, the established models were adopted to optimize process control parameters for spherical powder fabrication and predict powder properties. Validation experiments based on the agglomeration sintering powder preparation method demonstrate that the Gaussian process regression model achieves higher goodness-of-fit and lower prediction error for both prediction targets, and its predicted performance agrees well with experimental measurements. The proposed integrated experiment combining machine learning with powder material preparation and characterization can help enhance artificial intelligence literacy for students majoring in materials engineering.