Databases Reference
In-Depth Information
Figure 5.1
Confusion Matrix - Iris data
Beginning with the first objective,
Drag the model up to an available display and release.
Select “Confusion Matrix”.
The result is in Figure 5.1. In the title bar (top left) is the error rate. For this
classifier it is 0.7% - indicating that out of the 150 observations only one was
misclassified. The standard confusion matrix represents a N
N array of
observation counts or ratios where N is the cardinality of the output attribute.
The N rows represent the actual values in the data; the N columns represent
the predicted values. Correct classifications are represented by those down the
main diagonal. The VisMiner confusion matrix is plotted in 3-D and can be
rotated in the same way as other VisMiner visualizations. The Z axis represents
the count in each cell. The percentages are with respect to the total observa-
tions. The colors serve to distinguish between possible predicted/actual com-
binations. The error rate is the sum of all cells not on the main diagonal.
From the confusion matrix we see that the only misclassification was that a
single Virginica observation classified as Versicolor. Based on the extremely
low error rate, our classifier appears to do a good job.
When building classification models, it is a good idea to construct multiple
models using different classifiers, then compare the results. VisMiner supports
