The data requirements differ, depending on the approach taken to build the scoring card (manual or using the regression model) In the case of manual card building numerical as well as categorical attributes are allowed but missing values are not supported. To use data with missing values it is necessary to replace them before building the card. Otherwise all missing values will be assigned to the No Information level. In the case of building the card using the regression model, the data should satisfy the requirements of the logistic regression model (see the Logistic Regression chapter). In both cases the data set for constructing a scoring card should contain a binary target attribute.

There are two approaches to building scoring cards. It is possible to build the card 'manually' using only expert knowledge or on the basis of a logistic regression model. With the regression model approach the user obtains a scoring card with pre-calculated values for parameters such as points, being good and WoE.

In both approaches to scoring card building it is possible to edit all the parameters of the card: attributes, intervals and points. The user can add/remove attributes, edit levels, set his own points and define labels on the card.

In the case of the manual building process there are no restrictions on editing the card. The user can freely modify the card. One only has has to remember about saving the changes done to the card. In the case of building a scoring card with a regression model it is necessary to recalculate the points on the basis of regression model after editing the card. This can be done in the following way:

The new card with the recalculated points will appear in the repository.

More information about editing the cards can be found in the Model statistics section.

Additional usage

There is one additional very practical way of combining both approaches to scoring card building. In the case when the analyst needs to prepare several variants of the scoring cards it is possible to proceed in the following way:

• step 1:
  • manually build a basic scoring card ('BASIC_CARD')
  • work on the general appearance of the card: number of levels, level bounds, labels, descriptions, order of attributes, mapping e.t.c. Make the card easy to read, compare and evaluate. This card will serve as the template for the cards which will be created later.

• step 2:
  • build the final scoring card on the basis of a regression model
  • start the scoring card building process with the logistic regression model as described in the previous section
    • step 2a: build a logistic regression model ('MODEL')
    • step 2b: create a scoring card build task with following settings:
      • use 'MODEL' as regressionModel in the scoring card settings
      • use 'BASIC_CARD' as the inputModel in build task
      • set 'FINAL_CARD' as the name of the final scoring card

The picture below show the metadata repository after performing the steps described above.

Figure 40.7. Using a manually built scoring card as a template

After the first step the analyst can carry out the second step repeatedly, changing some settings in the regression model. In this way all the created final cards (changing as the logistic regression models change) will have the same appearance as the basic card. This will make it easier to e.g. compare the models and select the best one.

Note

• If the logistic regression model has more variables that basic card then the additional ones will be added to the final card.
• If the logistic regression model has fewer variables that basic card then the excessive ones will be removed from the final card.

Algorithm settings

Scoring card has the following algorithm settings:

Table 40.1.  Scoring card: General Algorithm Settings

Name	Description	Possible values	Default value
atPoints	The points level corresponding to odds for the PDO Scaling algorithm	positive integer numbers	300
Automatic Data Transformations	This option is ignored for the scoring card module.	-	-
Intervals from PhysicalData	if TRUE then the algorithm will use the intervals defined in PhysicalData (throught Freq), otherwise the intervals will be taken from function settings | algorithmSettings | attributesSettings	TRUE / FALSE	FALSE
Max Number of Points	The maximum value of points allowed for a level of an attribute. The upper bound for points scale. Used with the MaxPointsScaling algorithm	Positive integer numbers	100
Maximum For NoAnswer	If TRUE then the maximum from points of all levels will be ascribed to the additional No Answer level. Otherwise the minimum will be ascribed.	TRUE / FALSE	FALSE
Number of Intervals	The number of intervals created for an attribute.	Positive integer numbers	5
odds	The odds corresponding to atPoints in the PDO Scaling algorithm	positive floating point numbers	10
pdo	the number of score points which results in the doubling of odds used by the PDO Scaling algorithm
Swap Points	If TRUE the card will model the oposite category as the regression model, FALSE otherwise.	TRUE / FALSE	TRUE
Use Logical Data	If TRUE then the types of the attributes will be taken directly from logical data; otherwise they will be recognized automatically (i.e.. variables containing '__' in their names will be treated as categorical).	TRUE / FALSE	FALSE

Setting the number of intervals

To set number of intervals for each attribute open the attribute editor by double-clicking on the function settings | algorithmSettings | attributesSettings and set the required numbers (see picture below).

Figure 40.8. Setting the number of intervals

Model statistics

After creating the card the user can browse its details by double-clicking the model icon. The figure below presents the card editor and the card properties. The following animations show how to edit the card editing manual card, editing card with regression model.

Figure 40.9. Scoring Card Editor

Table 40.2. Scoring Card Model Properties

Name	Description
Liberal Execution	if TRUE 'liberal' execution is preferred (do not stop on minor errors)
negativeTargetValue	negative category value for the target attribute
positiveTargetApriori	the a priori probability of the occurrence of a positive category
positiveTargetValue	positive (event) category value for the target attribute
targetAttributeDataType	the type of the target variable
Target Attribute	the name of the target attribute
useApplyMapping	if TRUE applyMapping will be used during Apply and Test task execution; otherwise buildMapping will be used (the used mapping parameter is shown on the picture of Scoring Card Editor)

Note

The default values may differ depending on the card was built: manual or with a regression model. In the second case some parameters are taken from regression model settings.

Model testing is performed in the stardard way using the ClassificationTestTask regardless of how the card was built. It is important to set proper cut-off.

Figure 40.10. Model testing parameters

The scoring card model can be applied to classification problems.

The scoring card module is capable of creating two possible output types when classifying the provided data: category or probability. For details on how to apply the model to the data see the Applying Models in AdvancedMiner chapter, and the Classification subsection in the Applying for different mining functions section.

The table below presents the possible combinations and their descriptions.