集成SVM的回归预测及其遥感应用

Regression Prediction Using SVM Ensemble and Its Remote Sensing Application

为了提高单支持向量机(SVM)回归模型的泛化能力,引入遗传算法(GA)用以搜索SVM的"低偏差区域",给出了一种基于GA的SVM异构集成方法。用此方法对十个典型的数据集进行回归预测,并与单SVM回归结果和Bagging集成回归结果进行了比较,表明这种异构集成方法有较好的泛化能力。将这种方法应用于感兴趣的四个渭河陕西段水质参数的遥感反演,取得了更为精确的预测结果。实验表明,对小样本情况,基于GA的SVM异构集成方法能在付出合理时间花销的条件下,使单SVM的泛化能力得到有效提升。

To improve the generalization ability of a single SVM regression model,a heterogeneous ensemble approach of SVMs is proposed by employing GA to search low bias region.Ten representative data sets were regressed,comparing with the single SVM regression and Bagging regression,the heterogeneous ensemble approach has achieved stronger generalization ability.Applying this approach to regress four concerned water quality parameters of Wei River,more accurate results were obtained.The experiment show that,with rational time consuming,the GA-based heterogeneous ensemble of SVMs improves the generalization ability of the single SVM effectively when only a small data set is available.