When we say that the original features contain a lot of noise or redundant information, we mean that there is some information in the original data that is not of practical significance to the task we care about or that will interfere with our understanding and analysis of the task.
Let us understand through a popular example:
Suppose you are working on a task of predicting student performance. You collect some information about students, such as study hours, class attendance, family background, etc. This information is the characteristics.
Now, if these features contain some information that is not directly related to the student's performance, such as the student's clothing or the school lunch menu, then this information can be regarded as noise or redundant information.
-
Noise : refers to random or irregular information that has no practical significance for the task we care about. For example, in student performance prediction, students' clothing may not directly affect their performance, so this information can be regarded as noise.
-
Redundant information : refers to information that can be inferred from other features. They provide similar information to other features and make no additional contribution. For example, if you already have the student's study hours and subject grades, adding the number of tutoring classes attended each week may be redundant information because it provides similar information to the study hours.
During the data processing process, we usually take some methods to identify and eliminate this noise and redundant information to ensure that we focus on the truly task-related information when modeling and analyzing, thereby improving the performance and stability of the model. .