In order to evaluate the effect of military training of recruits scientifically, this paper discusses the problems of more qualitative evaluation, less quantitative evaluation and lack of scientific methods and evaluation standards. By analyzing the average training data of recruit classes, constructing projection pursuit model, using artificial fish swarm algorithm to solve the multi-constraint nonlinear optimization problem in the model, and using clustering analysis to cluster the results of eight training classes, the rationality, accuracy and reliability of the evaluation method are verified. The simulation results show that the algorithm model is efficient and easy to operate, and provides a scientific reference for the following intelligent military training.