A New Filter Method of Specific Sample Points Based on Partial Least-Squares Analysis

Specific sample point was a kind of noise that must be excluded from original data in the process of data mining and machine learning. A new filter method of specific sample points based on partial least-squares analysis was introduced in this paper. Two conceptions of true and false specific sample points were given and their relationship was elaborated in detail. The paper defined the critical value distinguishing true and false specific sample points and presented the critical formula. A novel method for identifying true and false specific sample points was described by using ellipse T² diagram, ellipsoid T² and scatter diagram of the principal component. Discrimination and filter method of specific sample points took great effect on eliminating samples created by random factors and purifying ultimate model.