Dimensionality Reduction Method of Training Sample Set for SVDD Based on Statistical Information
2012 ◽
Vol 220-223
◽
pp. 2097-2101
◽
Keyword(s):
In order to solve problem of high time complexity of support vector data description (SVDD) in training process, a low complexity SVDD algorithm is proposed by introducing statistical information grid (STING).The algorithm applies STING division to sample set space using support vector distribution characteristics and kernel distances between samples and sphere’s centre. Based on position information of sample points, it rejects non-support vectors and obtained simplified sample set. The results show that proposed method can reduce training scale and time without decreasing classification accuracy.