Question

I’m analysing a dataset with some data-mining tools.The response variable has ten levels and I’m trying to create a classifier.

Here comes the problem.When using nnet and bagging function,the result is not that good and the 5th level is even not in the prediction.

I want to use a confusion matrix to analyse the classifier.but as the 5th level is not shown in the prediction I can’t get a well-formed matrix.So how can I get a well-formed matrix?i.e. I want a 10*10 matrix.

The confusion matrix:

library("mda")#This is where **confusion** comes from
> confusion(pre.bag$class,CLASS)#here confusion acts like table
         true
predicted   1   2  3  4   6   7  8  9  10  5
       1  338   9  6  0   5  12 10  1  15 46
       2    9 549  1 59  18   0  3  0   0  6
       3   18   1 44  0   0   0  2  0   0  4
       4    0   1  0 21   0   0  0  0   0  0
       6    2  13  0  1 299   2  9  0   0  0
       7    5   2  1  0  10 231  6  0   1  0
       8    0   0  0  0   0   5 76  0   0  0
       9    5   1  0  0   0   0  0 62   0  0
       10   7   3  1  0   0   2  1  6 181 16
attr(,"error")
[1] 0.1231743
attr(,"mismatch")
[1] 0.03386642

Answer 1

Try this:

pred <- factor(pre.bag$class, levels=levels(CLASS) )
confusion(pre.bag$class, CLASS)

(Tested with an fda-object.)