I built a cross-validated SVM model with fitcsvm but I cant seem to use "predict" to predict responses from a new data with the model

Hi there

I built a cross validated and Bayesian Optimised model (CValidated_SVMModel_for_AUC1) with Support Vector Machines (SVM) on MATLAB with the "fitcsvm" syntax

I need to use the cross validated model to predict the responses from a new data (X_test)

I have tried to use this:

[Predicted_label,score] = kfoldPredict(CValidated_SVMModel_for_AUC1, X_test) without sucess

I have also tried to use:

[label,score] = predict(SVMModel_for_AUC1,X_test) without success

I will apprecaite if someone can advise me on how to resolve this

Thank You

[Adam Danz]

What problems did you have with the line below?

[label,score] = predict(SVMModel_for_AUC1,X_test)

[NCA]

sorry I meant to write

[label,score] = predict(CValidated_SVMModel_for_AUC1,X_test)

With this I get the error message

"No valid system or dataset was specified."

[Adam Danz]

How do you create CValidated_SVMModel_for_AUC1?

[NCA]

It was created with crossval syntax specifically CVPartition cross validation tool

CValidated_SVMModel_for_AUC1=crossval(SVMModel_for_AUC1,'CVPartition',c_part)

[Adam Danz]

In that case, it seems like kfoldPredict would the right tool. What error message or unexpected output are to getting with that? It would be easier to investigate this if you attach a mat file containing SVMModel_for_AUC1 and c_part.

Answer 1

[Muskan]

Hi

You can make sure that "CValidated_SVMModel_for_AUC1" is a cross-validated SVM model object created using "fitcsvm" and that it has been optimized and validated.

Cross-validated models contain multiple trained models (one for each fold). You need to extract the best trained model from the cross-validated object to use it for prediction. And you need to use the "predict" function with the extracted model.

You can refer to the following Stackoverflow answer for a better understanding:

https://stackoverflow.com/questions/55624612/no-valid-system-or-dataset-was-specified-while-using-predict-in-matlab

I hope this helps!