Abstract: In order to improve the classification accuracy of traditional nearest neighbor classification based on local mean and class mean（LMS）, it is necessary to study the weights allocation of class mean. The paper uses objective decision-making information to determine the weight of class mean as a mathematical formula. The simultaneous method with step size optimization techniques obtains a more appropriate distance weighting. A large amount of experiments by the UCI dataset show that the improved algorithm prevails over traditional LMS algorithm in classification effect.

Key words: class mean, step size optimization, weight, cross-validation

摘要： 为改进传统的基于局部均值与类均值的近邻分类算法的分类精度，有必要对分类中类均值向量的权重分配进行研究。将类均值向量的权重基于客观决策信息确定为数学公式，并运用步长优化的统一迭代法来对加权权重进行选取，通过UCI数据集大量实验表明，使用改进的分类算法进行仿真实验，结果表明，改进算法在正确率方面优于传统算法。

关键词: 类均值, 步长优化, 权重, 交叉验证