No, you cannot do it automatically. The labelling of ground truth data, including both training and testing data, has to be done manually by yourself or by the provider of a public dataset.

The classlabels has to be decided by you on the training dataset. Thereafter you can start testing using svm or any classifier...

Hi,

Ask yourself this question:

if I could label the examples automatically, then why would I need the svm?

Thank you all, but how to split so many features (extracted using detectSURFFeatures(), ExtractFeatures() in matlab) into training and testing sets.

You can write a piece of code to separate the total feature set into training and testing set depending on your need.

Yes that the point I need the code how am I goin to code in matlab

If there is problem writing code in matlab then it will be better if you take the feature file containing all the features in a separate folder and write a C/C++ file handing code to separate the total feature set into training and testing set depending on your need.

As mentioned in other answers, you have to first determine the two classes you have. Then you can assign each class a GT Label (+1 / -1).

Once you have your data and labels ready, you can randomly divide it into training and testing set using: datasample function in matlab without replacement. A rule of thumb for single fold validation is that your training set should contain 70% of the total data while the rest goes into testing set.

for example:

[~, idx] = datasample(trainingData, 0.7*numSamples, 2, 'Replace', false);

where numSamples is the number of total samples you have, 0.7 is taking 70% of the data.

This returns the idx which is the index to the selected elements in the training set. Everything else goes to testing set. I hope this answer is clear for you as it is my first attempt on this site.

Thank you I will try this.

I tried but I am getting Sample size K must be a nonnegative integer error.

the second argument is number of samples. When taking the percentage using 0.7, it does not correspond to an integer

Therefore you need to round it off to the nearest integer. try using ceil or floor to do this before you pass it to this function.