With the rapid development of Internet, cyber-attacks are increasing. Traditional security monitoring methods are inadequate to the massive data and potential attack risk prediction. This article proposes a method of machine learning based network attack warning and predictive protection, aiming to discover its own weak links through the use of the method, and automatically identify high adversarial attack behaviors from massive alarms, and adopt precise protection strategies in a predictive manner, thereby effectively blocking potential attack threats and improving network security protection capabilities.