Classification

In classification, the goal is to assign objects to categories/classes based on their characteristics. We distinguish the following types of classification problems:

• Binary Classification involves assigning an object to one of two possible classes. For example, if we want to determine whether a person has a specific disease or not, we are facing a binary classification problem.

• In contrast, Multiclass Classification involves assigning an object to one of three or more classes. For instance, if we try to categorize images of apples, pears, and oranges, identifying the correct fruit for each image, we are dealing with a multiclass classification problem.

• Multilabel Classification involves assigning an object to a subset of potential labels. For example, the label set could consist of topics for newspaper articles (e.g., politics, economy), and a newspaper article could discuss both politics and the economy simultaneously. In multilabel classification, it's essentially multiple binary classifications combined, which represent the multilabel output. The output is multidimensional, allowing for the recognition and exploitation of relationships between individual output components (labels).

• Multiclass Multioutput Classification looks at multiple classification problems together, similar to multilabel classification, but unlike multilabel classification, the individual classification problems may also be multiclass problems.

It is important to distinguish binary classification from multiclass classification, as some classification methods are specifically designed for problems with only two possible classes and cannot be directly applied to multiclass classification problems. The following table summarizes the terms introduced above.