I have a 20x20 confusion matrix. How can I calculate FAR, FRR, Accuracy and Precision values which represent complete system? Not for each class, but for the whole system.

有帮助吗?

解决方案

One simple approach might be to divide the sum of the diagonal elements by sum of all elements in the matrix.

That would be, for example, a generalisation of accuracy.

For example, precision might be calculated by taking the average of precision per class (PPC), which you get by dividing the diagonal element in a row by the row-sum and then averaging these PPCs for each row.

许可以下: CC-BY-SA归因
不隶属于 StackOverflow
scroll top