Evaluation index of multi-classification algorithm

In the past, most of the problems we encountered in solving classification problems were binary classification problems. Commonly used evaluation indexes include accuracy, precision, recall _ score, F 1 minute, ROC _ AUC _ score, etc. However, in real life, multi-classification problems also exist in large numbers. This section will explain the evaluation indicators of multi-classification problems in detail.

Let's take recall_score as an example to look at the evaluation indicators of two categories in sklearn library. In the recall_score method, there is a very important parameter' average', and its default value is' binary'. In the case of default parameters, this evaluation method can only solve two kinds of problems. If this evaluation method is used for multi-category problems, the system will report an error. But' average' also provides us with four other parameters' micro',' macro',' weight' and' sample' to solve the multi-classification problem. Let's take the iris data set as an example to explain these four parameters one by one.

[sk learn . metrics . recall _ score]('/p/59862986