Usage of PCA

Because PCA is a powerful technique for data transformation, you can apply it before further analytical work. This method is useful wherever high-dimensional data sets are encountered, especially in domains like text mining, image analysis, biological data analysis, customer preference, and taste analysis (collaborative filtering).

PCA is directly applicable to numerical data only because it is based on linear algebra calculations. Data sets that contain discrete attributes require a numeric encoding of their discrete values as a prerequisite. With all attributes being continuous, the input data set D can be represented by a matrix A. The rows of this matrix correspond to n attributes, whereas the columns correspond to |D| instances.

To obtain the eigenvectors (principal components) for matrix A, PCA applies a number of matrix operations. These operations comprise the transformation of attribute value vectors into a new representation space, also called feature vectors.