Which statistical measures for 4 class classification?

2
votes

I have a classification task with 4 classes which I solve with machine learning classifiers (SVM etc.).

Which statistical measures can be used for 4 classes? I will for sure use p-value (with permutation test) but I need some more.

Some interesting measures are true positive rate, true negative rate, positive predictive value, negative predictive value, are under ROC curve, are under the PR curve.... But I think that this are all only for 2 classes.

1
machine-learningstatisticsclassificationmultilabel-classification
You may find this interesting: stats.stackexchange.com/questions/51296/… - cel
I think what you're looking for is called the confusion matrix. - Robert Dodier

1 Answers

1
votes

There are several available metrics, described in the following paper:

Sokolova, Marina, and Guy Lapalme. "A systematic analysis of performance measures for classification tasks." Information Processing & Management 45.4 (2009): 427-437.

See Table 3 on page 4 (430) - it contains brief description and formula for 8 metrics; choose the most suitable ones for your task