Is it necessary to normalize a training data for KPCA?

I download some code from matlab file exchange. But nobody ensures about the data normalization that the data has zero-mean(approximately). The link to the code of matlab file exchange is here: https://www.mathworks.com/matlabcentral/fileexchange/39715-kernel-pca-and-pre-image-reconstruction

Answer 1

[Aditya]

Hi,

Yes, it is generally necessary to normalize your training data before applying Kernel Principal Component Analysis (KPCA). Normalization is an important preprocessing step for several reasons:

• Scale Sensivity
• Kernel Function behaviour
• Improved Convergence