Generate performance statistics for classification and regression algorithms in the
Model Statistics component. Use charts in the Model Statistics component to visualize model
performance results for Train, Validate and Test datasets and selected KPIs.
To configure the Model Statistics component to generate peformance statistics and
visualize results in the Model Statistics component, take the following steps:
- Launch Expert Analytics, connect to a dataset, and navigate to the Predict
Room.
- From the Component List add the Partition component.
Note You must add a Partition component if you want to see the charts when using Model Statistics.
The purpose of the charts is to display the curves for different
partition datasets.
- Double-click the Partition component, configure the required data
fields for the Train, Validate and Test datasets, and click
Done.
- In the Algorithms section, drag-and-drop selected algorithms to the analysis editor, and
configure them. For example, if solving a classification problem, you might
choose three classification algorithms, Auto
Classification, R-CNR Tree, and
Naïve Bayes.
- From the Component List add a Model Statistics component to the
analysis editor for the appropriate algorithm, regression or classifiction.
- Double-click the Model Statistics component to open the
configuration options.
- Click the Properties tab to configure the appropriate algorithm
type, set the Target Column on which to run the
algorithm, and then set the Predicted Column. Optionally,
click the General tab and add a component
Alias and Description. Click
Done.
- Click the Run Analysis
icon.
- Select the Results tab of Model Statistics component to see a
summary of the results.
- Optionally, view the data in the following chart formats:
- Gain/profit: Y-axis shows Gain/profit and X-axis
shows the percentage.
- Standardized (KS): Y-axis shows Standardized
profit and X-axis shows the percentage.
- Lift: Y-axis shows Lift profit and X-axis shows
the percentage.
- ROC: Y-axis shows Sensitivity and X-axis shows
specificity.
Note
For each chart, you can view Train, Validate and Test curves overlapping on the same
chart. There is one curve for each partition and for each chart.
You have generated performance statistics in summary and chart format for the required
alogorithm.