Learning Data Mining with Python - Second Edition

Using clustering algorithms as transformers

As a side note, one interesting property about the k-means algorithm (and any clustering algorithm) is that you can use it for feature reduction. There are many methods to reduce the number of features (or create new features to embed the dataset on), such as Principle Component Analysis, Latent Semantic Indexing, and many others. One issue with many of these algorithms is that they often need lots of computing power.

In the preceding example, the terms list had more than 14,000 entries in it—it is quite a large dataset. Our k-means algorithm transformed these into just six clusters. We can then create a dataset with a much lower number of features by taking the distance to each centroid as a feature.  ...