confidence score or confidence value of the classifier?

Hi

I am using computer vision for activity recognition .. I am dealing now with four classes using SVM classifier.

I need confidence score of SVM for each of the classes. is there any way used for computing the accuracy of the classifier??

I meant if it is possible to make the classifier make the prediction and show the confidence value of this prediction.

Thanks so much

[the cyclist]

What function(s) are you using to do the classification?

[Mammo Image]

for now I am using SVM classifier and planning to move to HMM.

any suggestion please? Thanks

[the cyclist]

I am asking specifically what MATLAB function(s) you are using. Even within MATLAB, there are different functions for some of these models.

Can you post your code?

[Mammo Image]

I am using fitcsvm function for two classes for now.

[Image Analyst]

What are the two classes? And what are your features? For example, I'm measuring weight, hair length, height, and vocal frequency range as features, and my two classes are male and female. It's not very meaningful to just throw a whole image at a classifier - you should pass it meaningful feature vectors.

[Mammo Image]

my current classes is sitting or standing , I am dealing with appearance features and I will deal with motion features. the classification process is OK, but I want to show any probability or confidence value with the result.

Thanks

[Mammo Image]

Image Analyst, any suggestion? still waiting your response.

thanks

[Image Analyst]

I'd try the Classification Learner app, or follow Star's advice.

[Mammo Image]

Thanks so much

Answer 1

[Star Strider]

Classifiers have their own special statistical properties. See the crosstab function.

[Mammo Image]

Thanks you, is it true that HMM shows the confidence value with the result of the output?

[Star Strider]

Apparently, for fitcsvm, there are two possible ways to calculate the probability: this version of kfoldpredict, and kfoldfun, neither of them obvious to me.

Explore these options and others to see what best meets your needs.

I have no experience with them since I’ve not used the SVM classifiers.

----------------------------------------------------------------------------------------------------------------------------------

EDIT — (14 May 2016 05:27 UTC) If you only have two classes, use the fishertest function. It will give you the confidence intervals.

[Mammo Image]

thank you very much I will try with them.

[Star Strider]

My pleasure.