粗糙集预测算法的稳定性分析

Stability of rough set-based prediction

粗糙集预测旨在从决策信息系统中学习规则从而预测新样本的标签.文中利用置信度刻画规则的可信程度,从而设计基于粗糙集的置信度预测算法,称为置信度算法.该算法可以对新样本分配与其匹配之后置信度最高的标签.泛化误差作为衡量算法有效性的指标之一,对其界的估计一直是构造学习模型的基础.利用算法稳定性概念刻画了置信度算法的泛化界,结果表明泛化能力由样本个数以及稳定性参数决定:样本数目越大,规则数目越多且稳定性参数越小;泛化误差界越小,经验误差越逼近泛化误差.

Rough set-based prediction aims to construct an algorithm to extract decision rules from decision systems and then assign the corresponding decision labels for new samples.To show the efficiency of prediction algorithms,it needs to measure the stability and generalization performance of algorithms.In this paper the generalization performance of confidence algorithm in terms of algorithmic stability is analyzed,which shows that the bound of generalization error is related to the number of rules,and stability parameter.The bound is decreasing as the increase of samples and rules,and the decrease of stability parameters.