Java: Data Science Made Easy

Unsupervised dimensionality reduction

The main idea behind feature extraction algorithms is that they take in some dataset with high dimensionality, process it, and return a dataset with much smaller set of new features.

Note that the returned features are new, they are extracted or learned from the data. But this extraction is performed in such a way that the new representation of data retains as much information from the original features as possible. In other words, it takes the data represented with old features, transforms it, and returns a new dataset with entirely new features.

There are many feature extraction algorithms for dimensionality reduction, including:

• Principal Component Analysis (PCA) and Singular Value Decomposition