Classifier scoring system

[Vispa]

Hello,

I am trying to use the custom classfier VI included in the Vision Development Module in 2010. I classifying three different types of images A, B and C and have created different type of feature fectors for them using different momentums and image processing techniques. However, I am able to get a separtaion for the three classes but I find it very confusing the way the classifier is scored. I need to understand it so that I can compare the different feature vectors that I have created and pick the best one. I have tried SVM and Nearest Neighbor classifiers and I also want to compare their performance.

Using the IMAQ Classifier Accuracy VI I am able to get some scores out. But I would appreciate if someone could explain more deeply the theoretical meaning of these scores:

- Class Predictive Value

- Classification Distribution

- Class Distance Table

- Accuracy

Thanks a lot!

[JtotheB]

Here is the help file on that.

Class Predictive Value;

Proportion of samples identified correctly per class.

Classification Distribution;

Table of classification results. Is this what you mean by class distance table?

Accuracy;

Proportion of samples classified correctly out of all samples.

If you have more questions, let me know. Hope that helps!