The influence of alkalinity (R), w(SiO₂) and w(MgO) on the granulation ratio, fresh pellets decrepitation temperature, compressive strength and falling strength of the fluxed pellets is studied by a pelletizing experiment, and a prediction and optimization model of the fresh pellet performance is established based on the BP neural network. The results show that, when preparing fluxed pellets, R is controlled between 0.6~1.4, w(SiO₂) is controlled between 3.5%~5.0%, w(MgO) is controlled between 1.8%~2.6%, and the resulting granulation ratio is above 90%. The increase of the proportion of fine-grained parts in the compound will increase the decrepitation temperature of the fresh pellets, increase the compressive strength first and then decrease, increase and then decrease the falling strength. The average absolute percentage errors (MAPE) of the model predicting the spherical granulation ratio, fresh pellets decrepitation temperature, compressive strength and falling strength are 2.64%, 5.09%, 5.75% and 11.19%, respectively, and the prediction accuracy could meet the requirements of the prediction error of the metallurgical industry and realize the accurate prediction of the performance of the fresh pellets. Combined with experimental research and model predictions, it is concluded that the performance of the fresh pellets is the best when the alkalinity of the compound is 1.0, w(SiO₂) is 5.5%, and w(MgO) is 1.8%.