偏最小二乘回归模型的泛化能力改进研究

On Improvement of Generalization Ability of Partial Least-Squares Regression Model

为了提高模型的泛化能力,提出了嵌入缩放思想的偏最小二乘回归(Partial Least-Squares Regression,PLS)建模方法。该方法通过对输入向量的缩放处理,将训练样本模糊化,缩小测试误差,从而提高了PLS的泛化能力。对原有的缩放法进行了改进,提出r算法。该算法可以找到合适的缩放因子,得到泛化能力更强的模型。仿真实验证明了所提方法的有效性。

An approach to improve the generalization ability of partial least squares regression （PLS）with shrink-magnifying thought is presented, which is implemented by shrinking or magnifying the input vector, reducing test error, improving the generalization ability of PLS. And r-algorithm is prsented based on shrinking-magnifying approach in order to find out the appropriate shrinking factor r and obtain better generalization ability of PLS. The simulation result shows the effectiveness of the proposed method.