As described in Chaps. 1 and 2 , the basic flow of pattern recognition is to convert patterns into feature vectors by feature extraction and then divide the learning patterns in the feature space into classes. However, the feature space and feature vectors, which should be called the original feature spaceOriginal feature space and the original feature vectors, Original feature vector cause a number of problems in classification processing if they are used as they are. To solve these problems, scaling, feature normalization, and feature selection represented by dimensionality reduction, are implemented. In this chapter, these are treated as feature space transformations in a unified manner and described one by one.