Module decomposition Copy item path

Matrix decomposition algorithms

§Dimension Reduction Methods

Dimension reduction is a popular approach for deriving a low-dimensional set of features from a large set of variables.

High Dimensional Data (a lot of input features) often degrade performance of machine learning algorithms due to curse of dimensionality. The more dimensions you have in a data set, the more difficult it becomes to predict certain quantities. While it seems that the more explanatory variables the better, when it comes to adding variables, the opposite is true. Each added variable results in an exponential decrease in predictive power. Therefore, it is often desirable to reduce the number of input features.

Dimension reduction is also used for the purposes of data visualization.

§References

• “An Introduction to Statistical Learning”, James G., Witten D., Hastie T., Tibshirani R., 6.3 Dimension Reduction Methods