Abstract: In order to overcome the uncertainty and subjectivity in deciding the evaluation index weight of battle scheme,linear and non-linear SVR-RFE are proposed. The proposed methods make weight vector and the cost function value to act as feature selection criteria,and SVR is employed to compare the regression ability. The simulation experimental results on a battle scheme samples show that the same performance can be achieved by linear and non-linear SVR-RFE methods. The optimum generalization performance is achieved when index feature is about 50% .

Key words: support vector regression, recursive feature elimination, evaluation index