Focusing on the problem of the basis functions selection for the linear-in-the-weights regression models, this
paper proposes a model structure optimization algorithm based on singular value decomposition(SVD) and PRESS(predicted
residual sums of squares) statistic. By dividing the original candidate matrix into several parts firstly, the comparison among
poor candidate regressors is avoided. Based on this operation, the SVD and PRESS statistic are applied to each sub-block for
candidate regressors selection. The generalization error of the model is taken as the direct target, and the candidate regressors
are selected in an adaptively manner. By using SVD, the number of the candidate regressors is reduced and the regressors
are orthogonal with each other, which effectively reduces the computation burden of the PRESS statistic. Simulation results
show the effectiveness of the proposed method.