The financial situation of China’s basic endowment insurance fund has begun to deteriorate, and its deterioration trend will accelerate with the deepening of the aging population, so it is urgent to carry out a study on the early warning of the payment risk of this endowment insurance fund. This paper discusses the classification accuracies of C4.5 algorithm, Naive Bayesian algorithm and BP neural network in the warning of basic endowment insurance fund payment risk. It is found that C4.5 algorithm has the best classification effect, with an accuracy of 71.43%; BP neural network takes the second place, with an accuracy of 61.90%; Naive Bayesian algorithm has a poor effect, with an accuracy of 52.38%.