Attribute weighting
The attribute weighting (English attribute selection or feature selection ) is also referred to as a sensitivity analysis . The English term indicates that they choose attributes based on whether they were relevant to the outcome of an experiment or decision-making process, and if so, to what extent.
Basic idea
In data mining records are often used as examples or instances (Engl. Instance , example ), respectively. They are characterized by a series of quantities called properties or attributes . In a decision-making process, the output data lead to a target variable which, in the simplest case, can assume two values and according to which the instance is classified . It is often interesting which of the attributes had which influence on the target variable, i.e. the class value of the instance. Finding this out is the goal of sensitivity analysis or attribute weighting. Their tools include the Relief algorithms , including ReliefF . To use them, it is first necessary to define a distance between the instances, which results from the differences between the attributes. Often the so-called Manhattan distance , the sum of the differences between the attribute values, is sufficient for this .
example
The following example is intended to provide an intuitive understanding of what is meant by the individual terms:
| Attributes: | outlook | temperature | humidity | windy | Class: | Gameday | |||||||
| possible Values: | sunny | cool | normal | No | Class value: | takes place | |||||||
| changeable | mild | high | Yes | was cancelled | |||||||||
| rainy | hot |
In the above example there are four attributes, two of which can have three values each, the other two attributes only two values. An instance is a concrete weather situation as a combination of the four attributes. By combining the attributes, different weather conditions can be represented in this example. Each instance can belong to one of two classes, the two possible class values of which are given by the decision whether a game takes place or fails under the weather conditions defined in the instance.
