Concept Drift Based on Subspace Learning for Intrusion Detection
Bin Wu, Hai-Zhuo Lin, Lin Feng · 2016
In recent years, Intrusion Detection System(IDS)thrives and becomes the main approach for detecting and defending internet attack.And network streams are the best data sources for studying network attack.In order to detect intrusions, concept drifting method is applied.What is more, the subspace learning based concept drifting method is fit for dealing with high dimensional data streams.It can not only detect the concept drift, but also reduce the dimensionality at the same time, which makes the detection more efficient.We also design model for judging concept drift, which checks the deviation of the error term of projection variance and the deviation of the error term of projection cosine.The experiment of KDD data set validates that our method is more efficient and accurate.