how do i clasiify non linearly separable data using unsupervised classification methods like k-means?

for example, iris data set, in which class 2 and 3 are not linearly separable as to class 1 which is linearly separable. Hence this often results in class 2 and 3 getting misclassified as 3 and 2 respectively.

what can be done in such a situation of non linearity between classes for iris and other similar data sets?

i am using k-means function and I am providing it with optimal cluster center's which i generated using global optimization technique.

Answer 1

[Shashank Prasanna]

KMEANS function in the Statistics Toolbox returns the 4 centeroids. You can compute the distance between an new point and each of the centrioids. The smallest distance will suggest that your data will belong to that cluster.

http://www.mathworks.com/help/stats/kmeans.html