I saw almost each regression method corresponds to a classification method. for example, adaboost classification, adaboost regression.
What is the relationship between them? Can I get a regression method from a classification method immediately?
$\begingroup$
$\endgroup$
4
-
$\begingroup$ What do you want to obtain from "regression"? To many people, in addition to estimates of a relationship among variables, they need quantitative assessments of uncertainty in those estimates: confidence intervals and the like. $\endgroup$– whuber ♦Oct 7, 2013 at 20:35
-
2$\begingroup$ Sometimes, a statistical model with qualitative response variable is called 'classification model' while a model with quantitative response is called 'regression model'. $\endgroup$– Michael MOct 7, 2013 at 20:40
-
$\begingroup$ It seems that each regression method usually corresponds to a classificaton method. for ex: support vector machine, support vector regression; adaboost, adaboost regression; I'm curious why so. $\endgroup$– user20756Oct 8, 2013 at 15:30
-
$\begingroup$ Sometimes, regression can also be used for classification, e.g. logistic regression. $\endgroup$– chaohuangOct 9, 2013 at 17:21
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Basically you use regression when your target is continuous and classification when it is categorical. Regression model make use of the total order of the target.
AdaBoost uses trees as base classifiers. In AdaBoost regression each leaf node of the tree is associated to a point estimation of a continuous random variable (the target), or even a full probability density function. Instead, in AdaBoost classification each leaf estimates a categorical variable.
Have a look at Decision Forest by Crimisini. There is one Chapter related to decision forests for regression and one for classification.
