Machine Learning with Spark - Second Edition

Interpreting the Eigenfaces

Looking at the preceding images, we can see that the PCA model has effectively extracted recurring patterns of variation, which represent various features of the facial images. Each principal component can, as with clustering models, be interpreted. Again, like clustering, it is not always straightforward to interpret precisely what each principal component represents.

We can see from these images that there appear to be some images that pick up directional factors (for example, images 6 and 9), some hone in on hair patterns (such as images 4, 5, 7, and 10), while others seem to be somewhat more related to facial features such as eyes, nose, and mouth (images 1, 7, and 9).