hello everyone,
I am trying to automatically classify different heartbeat classes using SVM. I have used many programs available on internet, but non of them works for me. I attached the training and test data and the program I used. could any one please tell me what is wrong with that? thank you in advance,
Any help is highly appreciated.
Hello Sir...Thank you very much for your help. can you please give me any clue of what format I need?
this is my Matlab code: (svm_tr and svm_ts are structures containing data and class number of each group in double format)
I have also attached my data. thank you very much. I really appreciate your help and consideration.
clc; clear all; close all;
data=load('svm_tr');
train_data=data.svm_tr;
data=load('svm_ts');
test_data=data.svm_ts;
xapp=train_data.data(300:654,:);
yapp=train_data.ClassNum(300:654);
SVMModel = fitcsvm(xapp,yapp,'KernelFunction','rbf','KernelScale','auto','IterationLimit' ,1e10);
%--------------------------------------------------
% testing
%--------------------------------------------------
Sample=test_data.data(3000:3828,:);
lable = predict(SVMModel,Sample);
acc=lable-test_data.ClassNum(3000:3828);
flag=0;
for i=1:size(acc)
if acc(i)==0
flag=flag+1;
end
end
test_accuracy=flag/size(acc,1)*100;
fprintf( '\nRate of correct class in test data : %2.2f \n',test_accuracy);
I have used just some part of data to check if it works for 2 classes. I am planning on expanding it to multi class after I get a good result with this one.
transposing seems a good idea I will definitely try it.
the links were really helpful. thank you.
thank you again for your time and consideration. that is really nice of you. I really appreciate it.
Hi,
There is this MATLAB toolbox called PRTOOLS
http://prtools.org/
I have used it myself, it has many classification routines, including muti-class SVM.
hope this helps
I have used LIBSVM package, it works well and is easy to use. You can download it from following link.
http://www.csie.ntu.edu.tw/~cjlin/libsvm/
Dear Shiva
I have seen your code . You need to implement multiclass SVM code. For that you have to follow the One Vs All SVM technique.
Eg If you have 5 classes for that traning data have alll data . and class label contain for class 1 positive and rest of the class negative.
for 2nd class, class 2 have positve label and rest will be negative.
and you have to train the usinf SVM funtion for 5 times and test also.
Means in your code your have to write svm function as many time as many class you have.
Hope this answer will be useful for you.
thank you all,
I am checking the feature extraction steps. I guess the problem is caused by one of the former steps that make data inseparable.
I will definitely consider your valuable advice. I really appreciate your help. thank you again.
- Badges
- Science topic