KPCA维数约简研究

Research on Dimensionality Reduction of KPCA

在对数据分析与处理时,为了避免高维数据所带来的巨大运算开销,通常需要对原始数据进行维数约简。与基于线性投影的维数约简方法相比,基于核方法的维数约简由于能够实现对样本的非线性映射,因此在数据预处理中具有更大的优势。对基于核方法的主成分分析(KPCA)维数约简方法进行研究,并通过实验结果证明KPCA不仅能够实现数据的降维,还具有增强数据线性可分性的优势。

To lower the heavy computation induced by high dimensional samples in data analysis, the original data is otten preprocessed lay dimension reduction methods. Due to the nonlinear map capability provided by kernel trick, the dimension reduction methods with kernel have a great- er advantage than those based on linear projection. Carries out a research on the Kernel Principle Component Analysis （KPCA）, and con- ducts an experiment on synthesis data. The experimental result shows that not only the dimension reduction but also linear separability can be achieved by KPCA.