Clustering process using SVM, unsupervised learning

Hello,

I am new in MATLAB. I have a large dataset (2+ millon points) containing 3 variables which I want to cluster/ classify into 3 groups based on the variation of those 3 variables. I have used K-means clustering method to cluster them. However, I was wondering is it possible to classify them using SVM? If yes, how should I move forward? Any suggestions will be appreciated. [Attched Sample Database matrix]

[Image Analyst]

I don't really see 3 classes here.

s = load('sample.mat')
sample = s.sample;
col1 = sample(1:10:end, 1);
col2 = sample(1:10:end, 2);
col3 = sample(1:10:end, 3);
plot3(col1, col2, col3, '.');
grid on;

[Mudasser Seraj]

They are not visually distinct classes. Rather closely related classes. The variables are time series data signifying human behavioural variations. And 3 classes represent normal, aggressive, defensive behaviour.

[Mudasser Seraj]

I have got the following type of clustering using k-means. Is it reasonable?

[Image Analyst]

Do you have any training or ground truth data? Do the classes correspond to the values you'd give to normal, aggressive, defensive behavior? Surely you must have some data where you've identified the behavior of the measurement set. If you don't, then how will you ever know if what it comes up with is correct?

[Image Analyst]

Do you have any training or ground truth data? Do the classes correspond to the values you'd give to normal, aggressive, defensive behavior? Surely you must have some data where you've identified the behavior of the measurement set. If you don't, then how will you ever know if what it comes up with is correct?

Have you tried classification learner to find out which method is best?

[Mudasser Seraj]

No. I don't have any ground truth data. I need to apply some unsupervised learning method to classify them. I was thinking of using multiclass SVM as an unsupervised learning method to classify the sample data. Is it possible?