0
$\begingroup$

I am trying to use clustering on my data but I do not have found the results I hoped for. I have a massive dataset with fire incidents. I would like to find clusters in these data. I want to use 4 categories to cluster the incidents. I would like to use the incident type, the response type, cause and the type of property. The result I am looking for is that the algorithm gives me clusters about types. Cluster 1: incidenttype A or incidenttype H, Responsetype X or Responsetype K , cause D and property R and so on. At this moment I tried to find this with K-means. But in the end I get clusters containing al most all categories of all four. Not nicely separated clusters. The first thought I have, is to check whether I use the best fitting algorithm?

$\endgroup$
  • 1
    $\begingroup$ It sounds like you don't want cluster analysis at all. If you know how you want things to come out, then you probably want some form of supervised learning $\endgroup$ – Peter Flom Nov 24 '16 at 14:31
  • $\begingroup$ How do you vectorize your dataset? Anyway Karan's is not performing very well if you have much noise or if your clusters are close to each other. I would use something like DBSCAN instead. It finds centers of high density and expand from then. I think it would work better for what you want $\endgroup$ – Vincent Teyssier Nov 30 '16 at 7:21
0
$\begingroup$

I played around with different methods, and decided to go for K-means. Played around with the settings and I get what I was hoping for. I filtered out the characteristics which do not happen very often (only a few times) and this how the incdent types in the data getting some more structural shapes.

It gave me pretty good results. I think I can get better results by finding the right amount of K clusters.

| cite | improve this answer | |
$\endgroup$

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.