soft question - Why and when is classification useful From What I understand about classification is that it is used to distinguish unlabeled data points in a collection. So if we have data which is labeled(age, height, weight,...) then we don't need classification, its used only in cases where data might be multidimensional but not classified, i.e without variables(age, height, weight,...)
Do I have the right understanding?
If not, what exactly is classification? and why and when is it useful?
Any help is much appreciated
 A: "Classification" is, in my view, a pretty broad term for any problem that involves sorting objects into discrete groups. 
These groups can be known in advance or not known. If they are known in advance, the task may be to develop a model that will be useful to predict future the class for future cases - one example of this would be models of disease, where we use cases with a known disease (or outcome) to develop models for patients who have symptoms but an unknown disease. In other cases, the purpose may be to explain why classes form. This explanation vs. prediction idea is also relevant in other areas (that is, where the problem is not classification). Techniques relevant here include various types of logistic regression and classification trees (not an exhaustive list). 
When the groups are not known, the problem may be to see whether groups even exist. Sometimes cluster analysis is used in this way. 
A: You misunderstand what is meant by "labels". You are talking about "labels" (names) for features/variables (height, weight, etc), but classification is talking about labels for the objects which have those features. The label is simply another feature or variable for the objects you're describing and in the future, when measurements of these objects come in with nothing in the "label" field, classification will attempt to fill that field in.
For example, you might be given height, weight, color, speed, eating (vegetarian, carnivore), etc, of a zoo-ful of animals. Some of them are identified as "zebra" or "kangaroo", etc -- they are labeled.
Supervised learning is where your algorithm has access to the measurements for each animal and also the label (species) that's been assigned to it. This data is called "training data" and your algorithm attempts to learn how to properly label future un-labeled animals based only on measurements. Unsupervised learning is where your algorithm has access only to the measurements (no labels) and tries to partition the animals into some kind of classes. Semi-supervised learning combines the two: your algorithm has access to the measurements and the labels for some of its training data, and also to a bunch of measurements without labels.
A: Classification is useful when you want to predict a discrete variable (e.g. an iris's species) given other variables (e.g. it's petal's width and length).
