Netflow data dimension reduction method with supervised discriminant manifold learning

The invention belongs to the field of data processing, and discloses a Netflow data dimension reduction method with supervised discriminant manifold learning. The method comprises the following stepsof: (1) obtaining a Netflow data packet from a local area network router and storing the Netflow data packet into a database; step (2) establishing a Netflow data matrix X = [y1, y2,..., yn]; (3) establishing a neighbor matrix Hij, and constructing a supervised discrimination matrix Sij in combination with the neighbor relationship of the neighbor matrix Hij; (4) calculating a local divergence matrix SL and a global divergence matrix SN; establishing a constraint objective function model J (A), and searching a low-dimensional projection subspace with a maximum global divergence matrix and a minimum local divergence matrix at the same time by utilizing the constraint objective function model J (A); and (5) feature decomposition: according to the constraint objective function model J (A), solving a solution of a con