预测缺失值

Question

如果我有下表：

A           B           C  
10,41090666 14,89846612 9,422344699
8,068361769 23,55384886 17,48139826
17,4985994  22,92848379 15,70903939
15,7719951  20,32010741 14,07175945
14,17758505 18,09037383 12,42757166
12,84353291 16,39917863 11,4492361
27,50382812 28,07940005 26,23678319
26,76107542 25,24077448 22,68518704
23,63261239 23,54045327 20,23789222
21,30484173 22,93639181 18,09231972
19,58502334 20,7761587  16,51749395
29,89083835 28,63481251 24,98733942
25,70124857 26,38312056 23,37880304
23,99115758 24,47022068 22,78210618
23,41206567 22,20480771 20,33867054
20,99705612 20,76698249 18,48344234
19,1719537  20,30674164 16,85224251
17,62578703 18,58890281 14,72958504
16,16422116 17,006585   14,50923107
14,51152313 15,58219258 14,08764833
13,82051007 14,28710942 13,87111687
22,11398149 19,95404742 11,83412051
19,77362546 18,75341386 18,80796142
18,2842237  18,42339932 x
17,10296288 16,39642442 x

如何预测缺失值？

Answer 1

整个task view on missing data。填写缺失值的一种简单方法是根据主成分分析进行单次估算。

library(missMDA)
ddi <- imputePCA(dd)$completeObs

证明估算的C值是合理的：

n <- nrow(dd)
cc <- rep(1:2,c(n-2,2))
pairs(ddi,gap=0,col=cc,pch=c(1,16)[cc])