Difference between PCA and KPCA
The transformed data obtained with PCA are called as Principal components.
This is obtained by calculating covariance between combination of variable.
ie, by performing covariance matrix.
[[Covariance(a, a), Covariance(a, b), Covariance(a, c), Covariance(a, d)],
We'll get N principal components from N-dimensional data. We might conclude
that the data size will not change after PCA, but the starting variables
will contain more information than the following variables. We can select
the P number of principal components from this based on
requirement. This allows us to minimize the data size while retaining as
much information as possible. This is done with the help of decision boundary.
The first line (blue line) is the one that has maximum variation and thus
the first principal component contains maximum information.
The second line (pink line) , on the other hand, has a bit less variance
than the first, and the information presented here is likewise a
little less. It is important to remember that the second line will be
unrelated to the first. This procedure is repeated until the desired number
of primary components has been reached (P).
(Note: P < N, P- dimension of data to be reduced ,
N is dimension of original data)
So, how this is done? This role is played by Eigen values and Eigen vectors.
The Eigen vectors contains the information about the direction of axes
which we call Principal components.
Eigen values contain the information about the amount of variance in each
principal component.
By ranking your eigenvectors in order of their eigenvalues, highest to
lowest, you get the principal components in order of significance.
KERNEL PRINCIPAL COMPONENT ANALYSIS
The main difference between PCA and KPCA is that, PCA can work well only in
linearly separable data.
For instance, we can say that PCA can only find the
best axis or decision boundary at which the data has maximum variation in a
linearly separable data; Whereas KPCA can work well in non-linearly separable
data too.
There is also another major difference, i.e. the covariance matrix in the
PCA is replaced with kernel matrix in KPCA.
If our objective is to perform classification or clustering of data,
then we can say that PCA is effective only if the data is linear.
If the data is not linearly separable like below figure, then we can go for
Kernel PCA.
Kernel PCA finds boundaries even if they are non linear functions
by mapping them into higher dimensional space.
Okay, How does this work ? Let's take a look at the below figure,
Application of kernel will map the data into higher dimensional space, which thus
makes the data linearly separable, but in a higher dimension.
This can be called as kernel trick.
We can calculate the Eigen value and Eigen vector directly
from this kernel matrix.
Thus reduced p dimensional data can be obtained
Detailed explanation about PCA here
Have a project enquiry? Do contact us
Good
ReplyDeleteNeat explanation
ReplyDeleteWell explained
ReplyDelete