0
$\begingroup$

I have 2 dimensional data about customer churn,i Clustered them with k-means but because of Outliers the clusters are not homogeneous.actually Outliers are about customers which are very important , so i can not ignore them,i used SOM method too,but again it didn’t work.please help me with this.

$\endgroup$
  • $\begingroup$ The goal of k-means, as an unsupervised machine learning method, is to generalize. In other words, it aims to learn the distribution of your data and generalize to new cases. This in turn means two things. First, you cant care about a single point which does not represent the whole data distribution. Second, you wish for the system to classify correctly for new cases, not the data you learned from. What you are trying to do is called over fitting in the generalization world. $\endgroup$ – havakok Oct 7 '18 at 12:44
  • 1
    $\begingroup$ Maybe k-means is not the suitable solution for you? It sounds to me as if you are not trying to generalize. If you are, maybe you can assign weights to certain more impotent data and apply a weighted k- means. In any case, more details on your final goal and a biger picture of the problem may help to better answer your question. $\endgroup$ – havakok Oct 7 '18 at 12:44
  • $\begingroup$ Do you have the header for your data? Imbalance in your features also matters but you are overfitting as previously was pointed out here. $\endgroup$ – HoofarLotusX Oct 7 '18 at 13:28
0
$\begingroup$

A solution can be apply clustering without outliers, and then if it is needed, apply new clustering method on the set of outliers.

Finally, you can unify the clusters of the method with each other as the final solution.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.