不同数据转换方法对特征数和聚类效果影响的研究

STUDY ON THE EFFECT OF DATA TRANSFORMATION METHODS ON CHARACTERISTIC NUMBERSAND RE SULTS OF CLUSTER ANALYSIS

本文在详细调查浙江省绍兴、德清和常山三县50个生态农业试验户的基础上,采用三级指标体系研究了原始数据和5种数据转换方法对指标平均值、最大最小值、标准差、标准误等待征数的影响及其对聚类分析效果的影响。结果表明各数据转换方法能使各特征数明显压缩、消除级差悬殊问题并使离散程度大大下降;用转换后的数据进行聚类分析,结果更为切合实际,而用原始数据直接聚类的效果很差。各种数据转换方法中,以(x_i-x_(min))/(x_(max)-x_(min))转换的效果为最好。

Based on detailed interviewed data of 50 ecoloigcal experimental farms in Shaoxing, Deqin and Changshan counties of Zhejiang province, effects of 5 data transformation methods and original data on averages, maximum and minimum values, standard deviations and standard errors of the indices at three hierarchical levels and on results of cluster analysis were studied. The results indicated that the characteristic numbers became much smaller, disparities of differences and discrete estent of the values among the indices effectively were reduced. Results of cluster analysis using transformed data were more effective and satisfactory than that using original data. Data transformation with OXi-Xmin/ (Xmax-Xmin) was most effective among the methods studied.