The act of reducing raw or high-dimensional data to a manageable set of representative features that capture the most important information for a certain job.
Definition of feature: Measurable property of a phenomenon being observed. If we have multiple features, we can define a rule.
Feature extraction is the act of reducing raw or high-dimensional data to a manageable set of representative features that capture the most important information for a certain job. In real applications usually many features are measured while only a small percentage of them carry useful information towards our learning goal. We usually need an algorithm that compress our feature vector and reduce its dimension.
Feature extraction is an important stage in machine learning since it helps reduce data dimensionality, remove noise or extraneous information, and emphasize the most discriminative elements of the data.
Curse of dimensionality states that higher dimensionality can lead to reduced classifier performance. It invites issues such as difficulty in visualization, increased computational complexity, overfitting and data sparsity.