Can anybody tell me what is the purpose of feature generation? and why feature space enrichment is needed before classifying an image? Is it a necessary step?
Is there any method to enrich feature space?
Data Science Stack Exchange is a question and answer site for Data science professionals, Machine Learning specialists, and those interested in learning more about the field. It only takes a minute to sign up.Sign up to join this community
Feature Generation -- This is the process of taking raw, unstructured data and defining features (i.e. variables) for potential use in your statistical analysis. For instance, in the case of text mining you may begin with a raw log of thousands of text messages (e.g. SMS, email, social network messages, etc) and generate features by removing low-value words (i.e. stopwords), using certain size blocks of words (i.e. n-grams) or applying other rules.
Feature Extraction -- After generating features, it is often necessary to test transformations of the original features and select a subset of this pool of potential original and derived features for use in your model (i.e. feature extraction and selection). Testing derived values is a common step because the data may contain important information which has a non-linear pattern or relationship with your outcome, thus the importance of the data element may only be apparent in its transformed state (e.g. higher order derivatives). Using too many features can result in multiply colinearity or otherwise confound statistical models, whereas extracting the minimum number of features to suit the purpose of your analysis follows the principal of parsimony.
Enhancing your feature space in this way is often a necessary step in classification of images or other data objects because the raw feature space is typically filled with an overwhelming amount of unstructured and irrelevant data that comprises what's often referred to as "noise" in the paradigm of a "signal" and "noise" (which is to say that some data has predictive value and other data does not). By enhancing the feature space you can better identify the important data which has predictive or other value in your analysis (i.e. the "signal") while removing confounding information (i.e. "noise").