A new minimax probability based classifier using fuzzy hyper-ellipsoid

In this paper, a new classifier called minimax-probability based fuzzy hyper-ellipsoid machine (MP-FHM) is proposed. It offers an alternative implementation of the minimax probability based classification with hyper plane and can be taken as an extended version of the ball-model based classifier. By the theorem proposed by Marshall and Qlkin, the training procedure of MP-FHM can be transformed into solving the corresponding unconstrained optimization problems, and thereby various optimization techniques can easily be adopted to solve them. In addition, the MP-FHM can be kernelized, and therefore it has strong nonlinear classification capabilities like other kernel-based classifiers. Various experiments were conducted and the results demonstrate that the proposed classifier is competitive with the state-of-the-art classifiers and is a very promising classification method.