Credit Scorecard Modeling Workflow

Create, model, and analyze credit scorecards as follows.

1. Use screenpredictors from Risk Management Toolbox™ to pare down a potentially large set of predictors to a subset that is most predictive of the credit score card response variable. Use this subset of predictors when creating the creditscorecard object.

   Create a creditscorecard object for credit scorecard analysis by specifying “training” data in table format. The training data, sometimes called the modeling view, is the result of multiple data preparation tasks (see About Credit Scorecards) that must be performed before creating a creditscorecard object.

   You can use optional input arguments for creditscorecard to specify scorecard properties such as the response variable and the GoodLabel. Perform some initial data exploration when the creditscorecard object is created, although data analysis is usually done in combination with data binning (see step 2). For more information and examples, see creditscorecard and step 1 in Case Study for a Credit Scorecard Analysis.

2. Create a creditscorecard object.

   Create a creditscorecard object for credit scorecard analysis by specifying “training” data in table format. The training data, sometimes called the modeling view, is the result of multiple data preparation tasks (see About Credit Scorecards) that must be performed before creating a creditscorecard object.

   You can use optional input arguments for creditscorecard to specify scorecard properties such as the response variable and the GoodLabel. Perform some initial data exploration when the creditscorecard object is created, although data analysis is usually done in combination with data binning (see step 2). For more information and examples, see creditscorecard and step 1 in Case Study for a Credit Scorecard Analysis.

3. Bin the data.

   Perform manual or automatic binning of the data loaded into the creditscorecard object.

   A common starting point is to apply automatic binning to all or selected variables using autobinning, report using bininfo, and visualize bin information with respect to bin counts and statistics or association measures such as Weight of Evidence (WOE) using plotbins. The bins can be modified or fine-tuned either manually using modifybins or with a different automatic binning algorithm using autobinning. Bins that show a close-to-linear trend in the WOE are frequently desired in the credit scorecard context.

   Alternatively, with Risk Management Toolbox, you can use the Binning Explorer app to interactively bin. The Binning Explorer enables you to interactively apply a binning algorithm and modify bins. For more information, see Binning Explorer.

   For more information and examples, see autobinning, modifybins, bininfo, and plotbins and step 2 in Case Study for a Credit Scorecard Analysis.

4. Fit a logistic regression model.

   Fit a logistic regression model to the WOE data from the creditscorecard object. The fitmodel function internally bins the training data, transforms it into WOE values, maps the response variable so that 'Good' is 1, and fits a linear logistic regression model.

   By default, fitmodel uses a stepwise procedure to determine which predictors should be in the model, but optional input arguments can also be used, for example, to fit a full model. For more information and examples, see fitmodel and step 3 in Case Study for a Credit Scorecard Analysis.

   Alternatively, you can apply equality, inequality, or bound constraints to fit a logistic regression model to the WOE data from the creditscorecard object using fitConstrainedModel.

5. Review and format credit scorecard points.

   After fitting the logistic model, use displaypoints to summarize the scorecard points. By default, the points are unscaled and come directly from the combination of Weight of Evidence (WOE) values and model coefficients.

   The formatpoints function lets you control scaling and rounding of scorecard points. For more information and examples, see displaypoints and formatpoints and step 4 in Case Study for a Credit Scorecard Analysis.

   Optionally, you can create a compact credit scorecard using

   To create a compactCreditScorecard object, use compact to create a compactCreditScorecard object. You can then use the following functions displaypoints, score, and probdefault from the Risk Management Toolbox with the compactCreditScorecard object..

6. Score the data.

   The score function computes the scores for the training data.

   An optional data input can also be passed to score, for example, validation data. The points per predictor for each customer are also provided as an optional output. For more information and examples, see score and step 5 in Case Study for a Credit Scorecard Analysis.

7. Calculate the probability of default for credit scorecard scores.

   The probdefault function to calculate the probability of default for training data.

   In addition, you can compute likelihood of default for a different dataset (for example, a validation data set) using the probdefault function. For more information and examples, see probdefault and step 6 in Case Study for a Credit Scorecard Analysis.

8. Validate the credit scorecard model.

   Use the validatemodel function to validate the quality of the credit scorecard model.

   You can obtain the Cumulative Accuracy Profile (CAP), Receiver Operating Characteristic (ROC), and Kolmogorov-Smirnov (KS) plots and statistics for a given dataset using the validatemodel function. For more information and examples, see validatemodel and step 7 in Case Study for a Credit Scorecard Analysis.

For an example of this workflow, see Case Study for a Credit Scorecard Analysis.