How to predict unknown data with a Regression​Partitione​dSVM-model

Hello,

Is it possible to use the result of fitrsvm(X,Y,'KFold',5) for predicting unknown, new data? If you have a normal RegressionSVM-model you can use the predict function for new data, but for RegressionPartitionedSVM-model you only have the kfoldpredict option, whose only input is the model itself.

Answer 1

[Lucio Cetto]

Hi,

Let me add to what Shubh already said, the partitioned model contains several models, so which one should be used for prediction? Partitioned models should only be used to asses the predictive performance of your modeling approach but not for pedicting on new data, it's better to train a new model using all the data once you are happy with the cross-validated results.

Lucio

Answer 2

[Shubh Sahu]

Hello,

There is no dedicated function which can predict unknown data with a regression Partitioned SVM model, but we are still working on it to enhance the forecast function so that it can be done. You can try these steps

mymodel=fitrsvm(X,Y,'KFold',5);  

yHat = kfoldPredict(mymodel)  

%estimate yHAt so that you can see the predicted value is approximately equal to observed 

You can write manually code for it. Here are some links that you can refer to:

https://www.mathworks.com/help/stats/regressionpartitionedkernel.kfoldpredict.html?s_tid=doc_ta

https://www.mathworks.com/help/stats/fitrsvm.html?s_tid=doc_ta

https://www.mathworks.com/help/econ/garch.forecast.html?s_tid=doc_ta