4
$\begingroup$

I am beginner to data mining. This is my understanding, regression is used to predict continuous values. It is a type of classification. Classification is Supervised and Clustering is unsupervised. In classification we have predefined classes, whereas in clustering there are no classes defined. What does this exactly mean? But in both of them we do have training data, which tells which output the set of features should belong to. Then how can we say that clustering does not have per-defined classes? Please explain when to use classification and clustering. Can clustering also predict continuous values?

Thanks. Please help.

This post has been edited.

$\endgroup$
  • $\begingroup$ Except for your last question, you will find useful clues by considering supervised vs. unsupervised learning. $\endgroup$ – chl Aug 21 '11 at 11:34
  • $\begingroup$ There are regression methods for all sorts of variables - logistic, Poisson, Cox PH, etc. $\endgroup$ – Peter Flom Aug 21 '11 at 11:40
  • $\begingroup$ @user904522 you don't need training data for clustering $\endgroup$ – Jeff Aug 21 '11 at 17:38
  • $\begingroup$ Thanks for the reply. So does that mean clustering never takes training data? $\endgroup$ – user904522 Aug 21 '11 at 20:27
  • 1
    $\begingroup$ By definition, unsupervised learning does not take training data, and clustering is usually considered unsupervised. However, you could use some hybrid technique where items are clustered without training data, but then those clusters are compared to a set of known data. If that's the case though, it likely means you have a pre-defined set of categories, and you might be better off using a supervised learning technique for classification. $\endgroup$ – Jeff Aug 21 '11 at 20:53
1
$\begingroup$

As @chl says, there are good threads on this site regarding supervised versus unsupervised learning. In regards to your bolded question: you're misunderstanding what data is supplied to supervised versus unsupervised methods. Supervised methods will always include an additional piece of information for each sample: the correct answer.

$\endgroup$
  • $\begingroup$ Thanks for the reply. So I understand from this that clustering never takes training data. Is it true? $\endgroup$ – user904522 Aug 21 '11 at 20:31
  • 1
    $\begingroup$ Yes, clustering algorithms are not trained: you don't run them on training data to have them adjust their parameters to maximize their chance of getting the correct answer (which is provided in supervised learning). But that doesn't mean that clustering algorithms don't have parameters that need to be chosen. You might use a set of data to figure out what algorithm and what parameters you intend to use in the future. (One example of a common parameter in clustering algorithms is the number of clusters it will seek to create.) This is not training and the data would not be called training data. $\endgroup$ – Wayne Aug 21 '11 at 23:33
1
$\begingroup$

Clustering: In clustering you group(cluster) the data based on some variables into some number of groups (cluster). Classification: In classification, you have certain groups & you want to know how different variables are related to the groups.

$\endgroup$
  • $\begingroup$ Thanks for the reply. So clustering never takes training data? If it does what does the training data contain? $\endgroup$ – user904522 Aug 21 '11 at 20:31
0
$\begingroup$

These video lecture's are really a good if you bring some time to listen to them. Particularly your questions about supervised vs. unsupervised should all be covered.

$\endgroup$

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.