Classification Wizard - Quality

In the "Quality" panel you can:

  • find out how accurately you can predict a state by looking at the labels in the matrix

  • check how important a variable (data point) is for the prediction model by using the importance matrix

  • test the classifications by entering (data point) input values for the classification test

Figure 1. Quality panel

The different parts of the Quality panel as well as how to use the Quality panel are described in more detail in the following.

Error Matrix

The Error matrix shows the data point classes that were created by labeling the data points in previous steps Classification Wizard - Clustering and Classification Wizard - Training.

The prediction model tells you how accurately you can predict a state. The error rate shows how accurate the prediction of a state is. In the example below, for example, the state "2-Check power supply" has the lowest error rate. Thus, the confusion matrix shows that the prediction for the state 2-Check power supply" is the most accurate one. The green highlighted columns contain the correct predictions and the other columns contain incorrect predictions.

Figure 2. Error Matrix

Importance Matrix

The importance matrix shows how important each data point variable is for the prediction model.

The importance matrix specifies how important a variable (a data point) is in order to predict a state. State means a labeled state such as 1-OK, 2 - Check power supply etc (see fígure below). Thus, the importance matrix specifies how important a variable is for the prediction model.

In the figure below you can see that the variable "PT Drive Voltage" is the most important variable to predict the states "OK", "2-Check power supply" and "Check ambient temperature" and the "Drive 1 Voltage" is the most important to predict the states "Check mechanical conditions".

Figure 3. Importance Matrix

Out of bag (OOB) Error rate

Error rate specifies the expected percentage of elements that cannot be classified correctly. This means the percentage of the data point elements that were selected for the prediction model and cannot be labeled.

Classification Test

Prediction

You can enter values into the table columns and then predict which classification (label) would be computed for these values. The available labels here are 1 - OK, 2 - Check power supply, 3 - Check mechanical conditions, 4 - Check ambient temperature.

Figure 4. Classification test

Result

The "Result" shows which classification would be computed for these values.