摘要
电类实验教学过程中人工评判学生所测数据工作烦琐,影响了教学质量和效率。该文提出了改进的K近邻(K-nearest neighbors,KNN)分类算法,即基于均值漂移、安全间隔和核主成分分析(KPCA)的M-KPCA-KNN(KNN based on margin and KPCA)算法,以判断学生测量数据正确与否和错误原因。首先利用KPCA对高维实验数据进行降维,然后利用均值漂移向量找到不同类别数据的最密集位置,并在不同类别数据的边界设置安全间隔,最后,将与待测样本距离最近的k个数据设置权重,计算每个类别的权重和,权重和最大的类别为待测样本的类别。与现有的KNN算法相比,M-KPCA-KNN算法不仅提高了分类正确率,而且降低了时间复杂度。
In the process of electrical experiment teaching,manual evaluation of the data measured by students is cumbersome,which affects the quality and efficiency of teaching.In this paper,an improved KNN(K-nearest neighbors)classification algorithm,namely M-KPCA-KNN based on mean shift,margin and KPCA(kernel principal component analysis)is proposed to judge whether the data measured by students is correct or not and the cause of the error.Firstly,the algorithm KPCA is uesd to reduce dimension number of the high-dimensional experimental data,then mean shift vector is adopted to find the densest locations of different types of data,and the safety margin among the boundary of different types of data is set.Finally,sets the weight of the k data points closest to the sample to be tested to calculate the weight sum of each category,and the category of the largest weight sum is regarded as the category of the sample to be tested.Compared with the existing KNN algorithms,the M-KPCA-KNN algorithm not only improves the classification accuracy,but also reduces the time complexity.
作者
申赞伟
周军盈
张士文
殳国华
张峰
SHEN Zanwei;ZHOU Junying;ZHANG Shiwen;SHU Guohua;ZHANG Feng(School of Electronic Information and Electrical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China;Shanghai Marine Electronic Equipment Research Institute,Shanghai 201108,China)
出处
《实验技术与管理》
CAS
北大核心
2022年第1期18-23,45,共7页
Experimental Technology and Management
基金
上海交通大学教育教学研究项目(JYJX200083)。
关键词
改进KNN
实验教学
均值漂移
安全间隔
核主成分分析
improved KNN
experimental teaching
mean shift
safety margin
kernel principal component analysis
