Scaling Techniques in Machine Learning

The technique of scaling generates infinite values on the objects to be measured. These techniques help understand the relationship between the objects. Let us see the techniques −

Comparative scales

It is the direct comparison of objects. The types include −

• Rank Order

• Constant sum scaling

Rank order

One item is judged against the rest of the objects. The respondents contain several objects who rank/order the objects based on a criteria. Rank order scaling is ordinal in nature, that is, (n-1) scaling decisions are made in this technique.

Constant sum scaling

In this technique, a constant sum of units is assigned to the respondent. For example, if a specific number of points indicate the importance of the product.

• If the attribute is unimportant, the respondent assigns 0 to it.

• If an attribute is twice as important as another attribute, it receives twice as many points.

The sum of all points is constant, that is 100, hence the scale name.

Non-comparative scales

In non-comparative scales, every object of the data set is scaled independently. The resulting data is assumed to be ratio scaled. The types include −

• Continuous rating scales

• Itemised rating scales

Continuous rating scales

It is a graphic continuum and generally has two coordinated extremes.

• Easy to construct.

• Simple to use.

• The respondent rates the object by placing a mark on a continuous line.

• The extreme values are not predefined.

Itemised rating scales

It is a graphic continuum and has two coordinated extremes.

• It is easy to use.

• It is easy to construct.

• The respondent rates the object based on a number or brief description associated with every category.

• The categories are ordered on scale position.

• Hence, the respondents pick the specific category that describes the object in question.