How to prepare the data for Classification learner Toolbox?

Hello

I would like to ask in respect of Classification Learner how many data sets should I have. Should I have data set for training and an other for testing or is it pssible to train on 10% of data and then test all population? Is it possible to do training and testing within toolbox window or I have to export model to workspace and test it separatly?

Thanks a lot!

Helen

Answer 1

[Bhargavi Maganuru]

You can refer Holdout Validation scheme , where you can select the percentage of data to use as a test set while training. But the final model is always trained using full data set. To explicitly test the model, you can export the model to Workspace and make predictions on new data. For more information about exporting the model refer this link

[EK]

Hello Bhargavi,

Thank you very much. If I set Holdout Validation scheme at 10% does it mean that I train classifier on 10% of my data and then apply at whole population?

When I export the model to workspace and make prediction on the new data I can not use the plots as in toolbox. Also the values look somehow odd. In Toolbox classifier gives me 85% accuracy. If I export model to workspace and apply it to the same data set the lamda values are near to 0