基于密度的覆盖算法

Research on Covering Algorithm Based on Density

覆盖算法分类时,一般是随机地选择样本作为领域中心进行覆盖。这种方法的不足之处在于,随机选择的样本将直接影响覆盖分类器的好坏,从而导致了分类器的不稳定性。通过提出一种基于密度的覆盖算法,试图根据样本的结构特征,通过结构最优化,每次有导师地选择领域的中心。这样既避免了覆盖算法的不稳定性,又使得每次训练出来的识别器,在结构上达到最优。通过试验与普通覆盖算法进行比较,结果表明该算法在一定的样本规模表现了分类正确性和稳定性上的优越性。

Covering algorithm selects a random sample as the center area to cover when classifying.The shortcoming of this method is that a random selection of samples will directly affect the quality of coverage classifier,which leads to the instability of classifier.This paper proposes a covering algorithm based on the density and there are instructors to choose center for the area according to the structure characteristics and structure optimization of the sample every time.Thus it can avoid the instability of the covering algorithm and make each trained recognizer optimal in structure.Through the experiment compared with the general covering algorithm,the results show the superiority on accuracy and stability of classification within a certain sample size.