Network threats are an obstacle to the protection of network and information security, and the current prediction of network security situation needs to be further improved. This study adopts the method based on artificial intelligence, and focuses on comparing the performance of Bayesian classification algorithm, SVM and RNN in the prediction of network security situation. Bayesian classification algorithm is based on the principle of probability statistics and can classify and predict network security events. RNN has the ability to process sequential data and capture temporal relationships, which is suitable for time-dependent modeling of network security data. SVM can handle small samples and noisy data, and has high stability and reliability in network security prediction. The experimental results show that the Bayes classification algorithm has the advantage of shorter execution time, which is suitable for the network security prediction scenario with high real-time requirement. RNNS, however, perform best in terms of prediction accuracy, and are better able to capture the temporal evolution trends and complex patterns of cybersecurity events.