缺失值問題在分類上的應用

Abstract

H探討缺失值問題在分類上的應用為主要目的。從參考過的文獻中，

禰◥滌眾]下，我們討論幾種缺失值處理法，包括(1)完整觀察值法、

值插補法、(3)迴歸法，以及(4)多重插補法，當運用在分類問題時，

□膆Ⅲ~率。此外，也探討當資料的缺失具單調型態時，不經插補進

A其理論錯分機率，並與上述四種方法處理結果比較。過程以電腦模

C

thesis, we investigate on the topic of classification with

values. After a review of the literature on missing values,

uss, under the assumption that all missing values are missing

om(MAR), various ways of handling missing values, including

lete observed vector method, (2)mean substitution method,

ession method, and (4)multiple imputation method. We compare

arent error rate(AER) resulting from the above four methods

plied to classification.

, suppose that the data with missing values can be arranged

notone pattern, we investigate the theoretical probability of

sification without imputing the missing values, and compare

ult to that of using the above four methods.

investigations are done by computer simulation.