Follow these steps to use the Propensity Scoring visualization.
Open a new workspace and click Add > Visualization > Predictive Analytics > Scoring > Propensity Score.
Set the Target (the dependent variable).
Set the dependent variable by selecting:
Dimension elements: Right-click in the workspace and select Table. Then select a Dimension elements as your dependent variable.
Filter Editor. Click Add > Visualization > Filter Editor to open the Filter Editor visualization.
After selecting a Dimension element or Filter as the dependent variable, click Set Target, enter a name to describe the dependent variable. Then click OK (and make sure the filter box is highlighted) to set the Target.
The name you give the target is the dependent variable that will appear in the left pane.
Add independent variables.
Add the independent variables using Metrics or Dimension Elements.
Metrics. From the Propensity Scoring toolbar, select a metric from the Metrics menu.
Dimension elements: Right-click in the workspace and select Table. Select one or more Dimension elements and drag to the left column under Independent Variables or to the Element box using the
Set Training Filter. You can define the set of visitors that you want to score by clicking Options > Set Training Filter from the Propensity Scoring toolbar. This will provide a subset of data built using only the visitors that you want to score. For example, who visited in the last month, visitors who reside in Australia, or visitors who viewed specific products.
The default filter is Train on Everyone, but you can change it by activating Dimension Elements in a table or building a filter using the Filter Editor.
After selecting a Dimension element or building a filter and while activated, click Options > Set Training Filter, enter a name to describe the filter, and then click OK.
Once you have identified all your inputs, press Go.
The scoring process will begin by passing over the data multiple times. It will then display the results as bar charts over a percentage line.
Save Propensity Score.
Starting with 6.1, you now have an option when using the Save Propensity Score:
Dimension and Metric
You can end up with two saved files, both a dimension and a defined metric.
If you submit the Propensity Score for processing you will get a dimension only.
The derived metric is the associated average score metric.
Check for accuracy.
The system will display Model Complete and generate a scoring model when the process is complete.
Right-clicking on Model Complete will identify the accuracy of the scoring model as defined by the system. Values ranging from 0 percent to 100 percent will identify the likelihood of the visitors matching the Target variable.
The Confusion Matrix gives four counts by the combination of Actual Positive (AP), Actual Negative (AN), Predicted Positive (PP), and Predicted Negative (PN). These numbers are obtained by applying the resulted scoring model to the 20% withheld testing data of which we know the true answer. If the score is greater than 50%, it is predicted as a positive case (matching the defined event).
|Accuracy|| Indicates how accurate the model is by identifying the correct predictions over all predictions.
(TP + TN)/(TP + FP + TN + FN)
|Recall|| Identifies the ability to re-identify the scoring model.
TP / (TP + FN)
|Precision||Identifies the level of discrepancy.
TP / (TP + FP)
Right-click on the Model Complete visualization and select Lift Chart, Gain Chart, or Model Viewer.