Rule Discovery System™ Modeling Edition

List of most important features

Prediction methods and model validation:

• All modeling methods can be applied to both classification and regression
• Recursive-partitioning (generates decision trees) for easy interpretation
• Covering (generates sets of independent rules) for easy interpretation
• Ensembles (Bagging, Random trees, etc) for maximum predictive performance
• Class weights can be used.
• Sampling of data can be used.
• All methods handle missing values.
• Class probabilities output in addition to predicted class for classification.
• Variable importance scores reported for all types of models.
• Validation methods:
  • random partition into training and test set
  • cross-validation (N-fold and leave X out)
  • explicitly grouped cross-validation
  • import of external test data
• Extensive statistics for model and method performance estimations
• Extensive statistics for model and method performance comparisons
• Easy to use any number of previously generated prediction models for prediction of new data.

Data manipulation:

• The modeling methods are such that there usually is no need for pre-processing of data.
• Independent variables (input) can be numeric, categoric, nominal and have lexical orderings.
• Dependent variable (output) can be categoric (classification) or numeric (regression).
• Data types are changed in a very simple manner.
• No limits in number of variables or examples that can be used (except primary memory of computer)

Model visualization module:

• Highly interactive browser for tree- and rule set- models. The user can quickly find the most informative rules.
• Variable importance histogram plots. Variable importance scores can also be exported.
• All visualizations can be printed, exported as images in a large variety of file formats, and copied to the clipboard.

Model performance visualizations:

• ROC-curves for classification models.
• Lift-charts for classification models.
• Predicted vs observed plots for regression models.
• All visualizations can be printed, exported as images in a large variety of file formats, and copied to the clipboard.

Data import and export:

• Data is read from text files and RDS own file formats.
• Data types are automatically guessed if not explicitly present in imported data.
• 'Intelligent' import wizard - user usually only needs to click OK.
• Export of RDS prediction models.
• Stored project files can be used for command-line scripting and running RDS in batch-mode.

Other:

• Command-line scripting available via XML-based protocol for input and output.
• The same XML-based protocol works as a simple and powerful API to RDS for application developers, in particular those who wish to make prediction models created with RDS available to a large number of users by integrating RDS Deployment Edition in any kind of IT infrastructure.
• Command-line version and RDS Modeling Edition and RDS Deployment Edition available on both Linux® and Microsoft® Windows®.
• All file formats used by RDS are portable between operating systems.