How to compute confusion matrix for cross-validated Naive-Bayes model?

1) Let mdlNB be a Naive-Bayes-classification-model. Then you can compute the confusion matrix as follows:

N=resubPredict(mdlNB)
[ldaResubCM,grpOrder]=confusionmat(resp,N)

2) Let mdlNBCV be a cross-validated-Naive-Bayes-Model (e.g.

mdlNBCV=crossval(mdlNB, 'CVPartition', cp)

)

Then the code above doesn't work:

NCV=resubPredict(mdlNBCV)

"Undefined function 'resubPredict' for input arguments of type 'classreg.learning.partition.classification.PartitionedModel'

How can I resolve this problem?

[Mihaela Jarema]

I think the code does not work, because mdlNB is a ClassificationNaiveBayes classifier, while mldNBCV is not a ClassificationNaiveBayes model, but a ClassificationPartitionedModel cross-validated, naive Bayes model, with a different set of methods. How about using another method instead, maybe kfoldPredict?

Answer 1

[Zuber Khan]

Hi,

You are facing this error because "mdlNBCV" is cross-validated classification model which means that it belongs to a set of classification models trained on cross-validated folds. For more information, you can refer to the following documentation:

https://www.mathworks.com/help/stats/classreg.learning.partition.classificationpartitionedmodel.html

As stated in the above documentation, in order to estimate the quality of classification by cross-validation, you should use KFOLD methods such as kfoldPredict, kfoldLoss, kfoldMargin, kfoldEdge, and kfoldfun.

On the other hand, resubPredict function classify data using a classification machine learning model, specified as a full classification model object. A list of supported models can be found here:

https://www.mathworks.com/help/stats/classificationsvm.resubpredict.html#:~:text=classification%20model%20object-,Classification,-machine%20learning%20model

I hope this answers your query.

Regards,

Zuber