How does the Classification Learner app calculate multi-class accuracy?

Hello,

I am using the classification learner app of Matlab R2022a in order to solve a 3-group classification problem.

After extracting the generated code and running it on Matlab, I get a confusion matrix from which I calculate indices such as accuracy, sensitivity etc. Since in a x-class problem you have x different true classes (each one at a time), I understand that you also have x measurements of accuracy, sensitivity etc. Nevertheless, the Classification Learner app shows a single classification accuracy. My question is: how does it calculate the universal accuracy? Is it from averaging the x accuracy measurements?

Answer 1

[Drew]

The doc page "Visualize and Assess Classifier Performance in Classification Learner" should answer all of your questions about classifer accuracy calculations in the Classification Learner app. On that page, in the "Model Metrics" section, the accuracy for a classification problem is defined as the "Percentage of observations that are correctly classified."