中文摘要：

提出一种最近邻分类的改良模型，综合考虑待分类数据的k近邻、所属的簇和整个训练数据集的类分布，充分利用局部、部分和全局三种类分布信息，从而具有抗噪声的性能。实验表明，提出的最近邻分类改良模型具有较好的抗噪声鲁棒性，而且分类的准确率明显高于传统的kNN分类算法。

英文摘要：

A classification algorithm is proposed based on mixture model of k-Nearest-Neighbor, Cluster and Data Set （NCD）. As the proposed algorithm takes the class distributions of the k nearest neighbors, the cluster and the entire training set into consideration, it can make full use of three kinds of classification information including local, partial and global information to achieve satisfying anti-noise performance. The experiment results show that in noisy environment, the NCD algorithm is significantly more accurate and more robust than the traditional kNN algorithm.