EL-SVDD: An Improved and Localized Multi-Class Classification Algorithm
Jian Xi Peng, Jian Xiong Tan · Applied Mechanics and Materials · 2015
In order to solve deviation and imbalance of the traditional multi-class classification. This paper designs an improved localized multi-class classification algorithm based on mutual communication entropy and Support Vector Data Description (SVDD), know as EL-SVDD algorithm. First, this algorithm calculates parameter values of the mutual communication entropy with many local classes of samples. Second, one class is placed inside the multidimensional sphere based on the mutual communication entropy. Finally, according to the samples and parameter values of the mutual communication entropy, it reinterpreted the C values of SVDD algorithm. As the result, the experiments shows that EL-SVDD algorithm not only has the feasibility, but also can improve the accuracy analysis of multi-class classification stably and effectively.