This paper describes a method for improving the generalization performance by means of the out-of-bag estimate for the generalization error in regression problems. We analyze the effect of the size of bags from the viewpoint of piecewise linear prediction achieved by the CAN2 (competitive associative net). Here, the CAN2 basically is a neural net for learning efficient piecewise linear approximation of nonlinear functions. We also examine and validate the effectiveness via numerical experiments.
抄録全体を表示