MATLAB: How to use Principal Component Analysis to reduce feature space

pca

I have collected 23 statistical features for each sample image(1 variable with dimension 800 by 23), to perform training with Neural network. I want to use Principal Component Analysis to reduce the numbers of features in my problem. when i get variance component from princomp function, it give a value of 99.1 for 1st componetnt/feature (for all examples). What does it mean? How can i use PCA to understand which features are good and which are not?

Best Answer

The ratio of 99.1 to sum(latent) indicates the proportion of the total variance contained in the first principal component. Depending on the problem, a threshold between 0.90 ad 0.99 is usually sufficient for determining how many components are sufficient for representing the original data.

Hope this helps.

Thank you for formally accepting my answer.

Greg

Related Solutions

MATLAB: Self-Organising Map (SOM) with Principle Component Analysis (PCA)

It is not clear if you have a well defined output.

If so, it IS NOT the variation of the inputs that are paramount.

It IS the variation of the outputs w.r.t. the inputs.

Check out principal COORDINATE analysis (very different from principal COMPONENT analysis!)

Hope that helps.

Thank you for formally accepting my answer

Greg

MATLAB: Principal Component Coefficients in princomp

The princomp function works on the centered data (variance/covariance matrix), so you need to remove the mean to reproduce the scores:

load hald
[a,b] = princomp(ingredients);
bsxfun(@minus,ingredients,mean(ingredients))*a

Best Answer

Related Solutions

MATLAB: Self-Organising Map (SOM) with Principle Component Analysis (PCA)

MATLAB: Principal Component Coefficients in princomp

Related Question