Description
Confusion Matrix Accuracy Precision Recall Precision or Recall F1 Score Sensitivity Specificity
Resources: Link to good website describing these
Evaluation metrics in practice
Having many evaluation metrics is hard to understand and optimise. Sometimes it is best to combine into one.
Use a single number i.e. accuracy or F1 Score .
This speeds up development of ml projects.
In order to use metrics to evaluate a model we can:
- Can combine multiple metrics a formula, i.e. weighted average.
- If there is a metrics we are happy that the model passes a given level then we can have it "Satisfying". So the for the given metric it just needs to pass a given level.
- For metrics we are interested in we have it "Optimising", the one we want to be the best.
- Setup: Pick N-1 satisfying and 1 optimising.
