Cluster analysis or regression?

I have a table with each row representing single printer model, its features, and price. I want to know how price is formed based on these features. What should I start with? Multiple regression, so I could cut off insignificant features? Cluster analysis to define small clusters with equal price? What are the ways to do the task?

• I think this depends on what you really hope to achieve with the analysis. In your case (given how you describe your data), both methods will be descriptive. Regression will help you answer a question such as which features have the strongest impact on price?, whereas clustering (such a MCA) will help you answer questions like which features are shared by printers of different price range?. Dec 21, 2012 at 12:28
• @AntoineVernet, that's really an answer -- and a good one. If you post it as an answer, you can earn reputation points. I will upvote. Dec 21, 2012 at 12:44
• @Placidia Thanks. I will make it an answer. However, I think PeterFlom answers the question better, I am merely stating that methods always depend on the question you are asking and that the first questions we ask from a dataset are usually too vague. Dec 21, 2012 at 14:10

Welcome to the site.

I don't see how cluster analysis helps you with what you want to do. Regression is much more appropriate. That is, you have a dependent variable (price) and a bunch of independent variables (features) = a classic regression problem.

Of course, problems may arise. This would depend on how many different printer models there are, how many features there are, how many levels each feature has, and so on.

• Sounds reasonable. I thought clusterization could say for example that there are two segments, 1)If a printer has wifi it will cost \$300 no matter what other features it has 2)All other printers where each feature is significant.
– Ivan
Dec 21, 2012 at 12:33
• +1, in addition, I might note that "cut[ing] off insignificant features" is unlikely to be a good idea. Dec 21, 2012 at 15:27
• @Ivan Those factors can be very important and ordinary regression cannot reveal them; but lots of plotting of the data can do so. After that, you may wish to fit splines or hockey stick models. Dec 21, 2012 at 18:01

I think this depends on what you really hope to achieve with the analysis. In your case (given how you describe your data), both methods will be descriptive. Regression will help you answer a question such as which features have the strongest impact on price?, whereas clustering (such a MCA) will help you answer questions like which features are shared by printers of different price range?