Note
that training with the augmented training set has diminished
classification accuracy for the white class by 41%.
How can
we fix this? Consider the confusion matrix...
|
Classification
|
|
True
Class
|
|
Asian |
Black |
Hispanic |
White |
| Asian |
0.73 |
0.19 |
0.00 |
0.08 |
| Black |
0.11 |
0.69 |
0.00 |
0.20 |
| Hispanic |
0.23 |
0.62 |
0.00 |
0.15 |
| White |
0.25 |
0.25 |
0.04 |
0.46 |
Note
there is a significant degree of confusion between the black and white
classes.
Let's
try adding a new feature to
sort this out...
