multiple feature for prediction time series data with libsvm

I am using libSVM in Matlab for prediction time series data. Basically, i use 1 feature (15 min past) to predict the label (the next 15 min value) and i got a good result from there.

data = x(1:end-1); dataLabels = x(2:end);

But, when i used multiple feature (ex: 4 --> 15 min, 30 min,45 min,60 min past data) to predict next value, it gimme an error.

v1 = x(1:end-4);
v2 = x(2:end-3);
v3 = x(3:end-2);
v4 = x(4:end-1);
label = x(5:end);
data = [v1,v2,v3,v4];
dataLabels = label;

      %splitting train test data
  trainDataLength   = round(length(data)*latih/100);
  trainSet          = data(1:trainDataLength);
  trainLabels       = dataLabels(1:trainDataLength);
  testSet           = data(trainDataLength+1:end);
  testLabels        = dataLabels(trainDataLength+1:end);

[MAPE,acc] = SVR(trainSet,trainLabels,testSet,testLabels,'nov-14');

Error says "Length of label vector does not match # of instances. model file should be a struct array". Then i am transpose the label.

SVR ran with 1 st iter but give no result with following message "label (1st argument) should be a vector (# of column is 1)." "Attempted to access predicted_label(:,1); index out of bounds because size(predicted_label)=[0,0]."

So, what should i do to solving this error ? Thanks in advance :)

Answer 1

[Stalin Samuel]

Multi-Class SVM

[agri kridanto]

Thanks for your answer sir, But, i thought that solution is for multiclass classification using libsvm. In that solution using 4 features in fisheriris dataset and class provided. i can run the solution without any problem, but in my case for regression which has time series data. I got the result with 1 feature (15min past data) and the label is next data. But i cant use 4 (60,45,30,15 min past data) feature as in my question.

Am i misunderstood your solution, sir ?

Answer 2

[lemon Nation]

I am in the same difficulty as you，I think there may be something wrong with the setting of the tag

[lemon Nation]

Just now this problem has been solved by me, sharing methods in the following hope can help you

Wrong codes:

clear all,clc

format compact;

nn = 5, n=10

x = -2.5:1.5

err=x.^2

model = svmtrain(err,x,'-s 3 -t 2 -c 2.2 -g 2.8 -p 0.01')

[py,accuracy,dv] = svmpredict(err,x,model)

%%

testx = nn+1:n

testy = zeros(10,1);% 理论y值无所谓

[ptesty,~,~] = svmpredict(testy,testx,model)

err = [py,ptesty]'

%% 出现如下问题：

>>label (1st argument) should be a vector (# of column is 1).

corrected codes

clear all,clc

format compact;

nn = 5, n=10

x = [-2.5:1.5]'

err=x.^2

model = svmtrain(err,x,'-s 3 -t 2 -c 2.2 -g 2.8 -p 0.01')

[py,accuracy,dv] = svmpredict(err,x,model)

%%

testx = nn+1:n

testy = zeros(10,1);% 理论y值无所谓

[ptesty,~,~] = svmpredict(testy,testx,model)

err = [py,ptesty]'

%% Tips

This is the first time answer others' questions and hope I hope I can keep the good habit of keeping records