Ensemble Methods in Machine Learning-英文文献.pdf

Ensemble Metho ds in Machine Learning Thomas G Dietterich Oregon State University Corvallis Oregon USA tgdcsorstedu WWW home page httpwwwcsorstedutgd Abstract Ensemble metho ds are learning algorithms that construct a set of classiers and then classify new data p oints by taking a weighted vote of their predictions The original ensemble metho d is Bayesian aver aging but more recent algorithms include errorcorrecting output co ding Bagging and b o osting This pap er reviews these metho ds and explains why ensembles can often p erform b etter than any single classier Some previous studies comparing ensemble metho ds are reviewed and some new exp eriments are presented to uncover the reasons that Adab o ost do es not overt rapidly Intro duction Consider the standard supervised learning problem A learning programis given training examples of the form f x y x y g for some unknown func m m tion y f x The x i values are typically vectorsof the form hx x x i i i in whose components are discrete or realvalued such as height weight color age and so on These are also called the fe atures of x i Let us use the notation x ij to refer to the j th feature of x i In some situations we will drop the i subscript when it is implied by the context The y values are typically drawn from a discrete set of classes f K g in the case of classication or from the real line in the case of regr ession In this chapter we will consider only classication The training examples may be corrupted by some random noise Given a set S of training examples a l