Multiple Birth Support Vector Machine with Triplet Loss Function.

TWSVM which can be regarded as an efficient binary classification algorithm has achieved much attention. However, when it comes to multi-class classification, scholars have to consider other algorithms. Therefore, many multi-class TWSVM has been proposed to deal with multi-class problems. In this paper, we use multiple birth support vector machines (MBSVM) which is an efficient algorithm for multi-class classification in which the decision criterion is the farthest distance of the test pattern to the hyper-planes, rather than the closest distance in multi-class TWSVM. The algorithm has much lower computational complexity and can be expected to be faster than the existing multi-class SVMs. However, when facing multi-class problem of imbalanced data, the MBSVM which adopts hinge loss is easily leads to instability for resampling. To enhance the performance of the MBSVM, we present a novel MBSVM with the triplet loss (tMBSVM) which deals with the imbalanced dataset problems and shows differences between positive data and negative data in one class. Numerical experiments on data sets demonstrate the feasibility and validity of our proposed method.