基于偏最小二乘特征提取的支持向量机回归算法

A Support Vector Machine Regression Algorithm based on Partial Least Squares Feature Extraction

为了提高SVM的建模质量,简化建模难度,提出了PLS-SVM组合回归建模方法。该方法通过PLS对样本数据进行降维、去噪以及消除共线性处理后,再进行SVM回归建模。不仅保持了SVM良好的模型性能,而且使SVM具备特征提取功能。实验结果表明,该方法是可行的,采用此法构建的SVM模型,泛化性能优于没有特征提取的SVM。

A hybrid PLS-SVM method is proposed to improve the SVM model quality and reduce the modeling difficulty. Firstly, reduced the dimensions of correlated inputs and denoised for sample by PLS, then construct the SVM model. The PLS-SVM not only maintains the SVM good performance but also has feature extraction function. The experiment results show that this method is workable well and the generalization ability of SVM with feature extraction using PLS is much better than that without feature extraction.