Building a Credit Scoring Model

Credit Scoring

Credit scoring is a technique used to determine whether or not to extend credit (and if so, how much) to a borrower. This workflow illustrates how to create and choose a credit scoring model based on both historical data and on the application of different machine learning algorithms.

Data Reading

The data are German Credit data, including credit status, demographic data, and customer history. The file is located in TheData/Credit.

Pre-processing

Learners such as neural network or SVM (Support Vector Machines) can only handle numeric attributes.

Nominal columns are converted into numerical columns.

Model Training and Evaluation

The following algorithms are trained and evaluated with cross-validation:

1. Neural Network

2. SVM (Support Vector Machines)

3. Decision Tree

Double-click on the metanode to see the subworkflow

Model Selection

All results, i.e. accuracies and respective models, are combined in one single table.

Rows are then sorted by descending accuracy and only first row (best performing model) is kept.

Save the Model

• Convert the "model" cell back to PMML

• Save the model

KNIME Analytics Platform writes out the model in the official PMML format, so that other applications can use the model.

Visualize

Compare accuracy scores of the three models.

Task

Create a credit scoring model based on historical data. Select the best machine learning algorithm to be applied. Use cross-validation to evaluate model performance.