Hi Esther,

SVC class weighting is an interesting approach building in prior knowledge about class imbalance during the optimization.

It’s trying to address the same problem as ROC however inside the algorithm, not outside as ROC.

It is now supported in sdsvc (unreleased). Find attached sdsvc.p and sdsvc.m files.

With ‘w’ option, you may provide a vector of class weights in the same order as the list of the input data set. It is supported for two-class and one-against-one multi-class option for all RBF, poly and linear SVM. Automatic meta-parameter selection is supported as well.

The weights multiply given C (they do not need to be in <0,1> range.

To me, it’s not really clear how to fix the weights to get a desired outcome. Generally, it’s good to lower weight on large classes and increase on important small ones.

```
>> b
```

1080 by 2 sddata, 3 classes: 'A'(1000) 'B'(40) 'C'(40)

>> p1=sdsvc(b,'sigma',1,'c',1,'one-against-one')

one-against-one: [1 of 3: 'A' vs 'B' SVs=141]

[2 of 3: 'A' vs 'C' SVs=130]

[3 of 3: 'B' vs 'C' SVs=37]

sequential pipeline 2x1 'Scaling+SVM stack+Multi-class combiner+Decision'

1 Scaling 2x2 standardization

2 SVM stack 2x3 3 classifiers in 2D space

3 Multi-class combiner 3x3

4 Decision 3x1 weighting, 3 classes

>> sdconfmat(b.lab,b*p1,'norm')

ans =

True | Decisions

Labels | A B C | Totals

-----------------------------------------------

A | 0.995 0.002 0.003 | 1.00

B | 0.850 0.150 0.000 | 1.00

C | 0.675 0.000 0.325 | 1.00

-----------------------------------------------

>> p2=sdsvc(b,'sigma',1,'c',1,'one-against-one','w',[0.01 0.9 0.9])

one-against-one: [1 of 3: 'A' vs 'B' SVs=785]

[2 of 3: 'A' vs 'C' SVs=810]

[3 of 3: 'B' vs 'C' SVs=37]

sequential pipeline 2x1 'Scaling+SVM stack+Multi-class combiner+Decision'

1 Scaling 2x2 standardization

2 SVM stack 2x3 3 classifiers in 2D space

3 Multi-class combiner 3x3

4 Decision 3x1 weighting, 3 classes

>> sdconfmat(b.lab,b*p2,'norm')

ans =

True | Decisions

Labels | A B C | Totals

-----------------------------------------------

A | 0.595 0.217 0.188 | 1.00

B | 0.000 0.925 0.075 | 1.00

C | 0.000 0.200 0.800 | 1.00

Does it work for you?

Best,

Pavel

sdsvc.p (File Size: 5KB - Downloads: 38)

sdsvc.m (File Size: 4KB - Downloads: 34)