Input
iris
pid |
prob1 |
prob2 |
obs |
2 |
0.66667 |
0.33333 |
1 |
4 |
0.66667 |
0.33333 |
1 |
6 |
0.66667 |
0.33333 |
1 |
8 |
0.66667 |
0.33333 |
1 |
10 |
0.66667 |
0.33333 |
1 |
12 |
0.66667 |
0.33333 |
1 |
14 |
0.66667 |
0.33333 |
1 |
16 |
0.66667 |
0.33333 |
1 |
18 |
0.66667 |
0.33333 |
1 |
20 |
0.66667 |
0.33333 |
1 |
... |
... |
... |
... |
SQL Call
Because this call specifies AUC ('true') and Gini ('true') and omits the ROCValues argument, the ROCValues argument has the value 'false'.
SELECT * FROM ROC (
ON iris AS InputTable
OUT TABLE OutputTable (irisout18)
USING
ProbabilityColumn ('prob2')
ObsColumn ('obs')
PositiveClass ('2')
NumThreshold (100)
AUC ('true')
Gini ('true')
) AS dt;
Output
irisout18
auc |
gini |
0.9464
|
0.8928 |
The AUC value is much greater than 0.5, which means that the model performs well, and you can use it for prediction.