Principal Component Analysis
As seen in the image below, PCA explores the example space <math>X</math> to find the maximum spread of data (variance). This will determine the first feature (big arrow). Then it collapses the example space into the hyperplane that is orthogonal to the feature vector, where dimension is <math>N-1</math>, and proceeds iteratively until there are no more dimensions. As a result, we get an orthogonal base that is generative of the original hyperspace.
The resulting vectors are ranked from big to small variance, so the pruning process starts from the last one found, the one with least variance. In the image below we would start removing the smallest vector in the left.
PCA will not work with data that is not of the same nature. For example,