With the rapid development of information technology, the problem of network security has become increasingly prominent. Camouflage intrusion, as a common means of network attack, has strong concealment and destructiveness, which brings great security threats to enterprises and organizations. In order to effectively deal with camouflage intrusion, more and more researchers apply machine learning and data mining technology to the field of intrusion detection. Among them, Random Forest (RF) algorithm, as an ensemble learning algorithm, has the advantages of high accuracy and low complexity, and has been widely concerned. However, the traditional RF algorithm still has some problems when dealing with camouflage intrusion detection, such as single feature selection, strong correlation between base classifiers and so on.