Research on Network Behavior Risk Measurement Method Based on Traffic Analysis

Table 3

Comparison of dimension reduction algorithms.

Algorithms	Advantage	Disadvantage
PCA	Linear mapping to eliminate the interaction of data	Only deal with the sample variance, not nonlinear data
LDA	Used for big data classification	Not suitable for non-Gaussian distribution samples
KPCA	Nonlinear data can be processed on the basis of PCA	Depend on the choice of kernel function
MDS	Keep sample difference on Euclidean distance	Not consider the distribution and interaction of adjacent data
Isomap	Preserve the geometric properties of samples on manifold distance	Not suitable for manifolds with large curvature
LLE	Solve the problem of high-dimensional data distribution	Need to assume that the manifold of the sample exists