Abstract:

Compared with the traditional support vector machine, the twin support vector machine owns faster calculation speed. However, the twin support vector machine is easy to produce over-fitting and has low computational efficiency, which does not have structure risk minimization characteristics and parsimoniousness. In order to solve this problem, an improved sparse twin support vector regression (ISTSVR) algorithm is proposed. The twin support vector regression algorithm combined with structure risk minimization principle improves the regression performance of the algorithm by adding a canonical term in the objective function. At the same time, a subset of train samples is selected to take place of the whole train sample, which makes the kernel function from square into a rectangular matrix, thus making the algorithm own sparseness and effectively reduces the computation time. Simulation results are provided to validate the effectiveness of the proposed algorithm.