Unequal costThis refers to the situation where different costs of losses are assigned to each category.

In some cases, different prediction results may cause different degrees of loss. It is not practical to assign the same cost function to such classifications. In this case, "unequal costs" can be assigned to errors based on the importance of the classification category.

Under unequal costs, what is considered is not the number of errors, but minimizing the "overall cost".

When considering unequal cost situations, the conventional metric ROC curve cannot represent the situation of the learner. Usually, a "cost curve" is used to illustrate the expected overall cost of the learner.

Parent word: cost

References:

Machine Learning, by Zhou Zhihua, Tsinghua University Press