Decomposition of Generalization Error for Weighting Fused Estimator

The performance of machine learning may be expressed by the generalization error. The less generalization error is, the better the performance of machine learning and on the contrary, the worse. To further study characteristic of machine learning algorithm, the decomposition method of the generalization error for estimators is usually used. Wherein the bias-variance decomposition for quadratic error loss functions is well known and serves as an important tool for analyzing supervised learning algorithms. In this paper, Aiming at the weighting fusion method and quadratic error loss function, we give detailed decomposition process of generalization error for weighting fused ensemble estimators. On the basis of this, the decomposition equation for the optimal fused method is further obtained.