Hello!! I have a vector of size "1*4006". The values are decimal numbers. These values are basically MFCC features. I want to apply multi-class SVM on it. My output will have 3 classes, that is, [0,1,2]. Someone please help me to do its code. I am n

Hello!! I have a vector of size "1*4006". The values are decimal numbers. These values are basically MFCC features. I want to apply multi-class SVM on it. My output will have 3 classes, that is, [0,1,2]. Someone please help me to do its code. I am new in this field. I have tried svmtrain function but i am not able to give input parameters properly.

thanks in advance

Answer 1

[Walter Roberson]

As discussed in https://www.mathworks.com/matlabcentral/answers/385076-hi-i-am-using-2017a-matlab-i-am-not-able-to-use-libsvm-on-it-the-image-shows-the-error-should#comment_540038 svm cannot be meaningfully used when you do not know which class each sample belongs to.

... Unless, that is, you want to investigate techniques such as https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2099486/ which you would probably have to write your own code for.

[SHWETA KHARA]

hey I know to which class each input belongs to. All I am stuck at point is I am not getting how to give input to svm..

[Walter Roberson]

As discussed in https://www.mathworks.com/matlabcentral/answers/385076-hi-i-am-using-2017a-matlab-i-am-not-able-to-use-libsvm-on-it-the-image-shows-the-error-should#comment_540133, the tool you are using expects that one of the columns contains the output that is appropriate for that that input row -- expects that one of the columns will contain 0, 1, 2, or 3 for that row. You would mark that one row as Response and leave the other rows as Predictor.

If you have a separate variable that contains the class then build a new matrix of the samples concatenated with the known targets and tell the tool you are using which column of the combined matrix contains the targets.