KnowledgeSTUDIO provides advanced predictive modeling capabilities such as linear and logistic regression, neural networks, scorecards and market basket analysis—appealing to a wide range of business and expert users.
Business users benefit from an intuitive wizard-driven model building interface with intelligent defaults—no statistical programming skills are required.
Quantitative analysts have access to more complex functionality with configurable settings for fine tuning advanced model parameters to accommodate models of any complexity.
KnowledgeSTUDIO provides common features for regression and neural network models:
- 4 standard link functions – Logistic, Probit, Log-Log and Complementary Log-Log
- Generation of scores and cross-validation
- Automatic translation of models to SAS, PMML and XML code
- Support for weights
| Regression Models |
Feature Highlights |
| Linear and logistic regression models including multinomial logistic regression |
- Automatic conversion of categorical variables to a set of indicator variables
- Variable selection methods – forward, backward, stepwise, R2 and user-defined order—with the option to force variables into the model
- Automatic calculation of Variance Inflation Factors
- Automatic checking of collinear variables and linearly separable variables (in logistic regression)
- Parameter optimization algorithms – Newton-Raphson and Levenberg-Marquardt
|
| Neural Networks |
Feature Highlights |
| Linear and logistic regression models including multinomial logistic regression |
- User-defined number of hidden layers and number of neurons per layer
- Automatic normalization of variables and conversion of nominal variables to indicator variables
- Parameter optimization algorithms – BFGS, Classic Back-Propagation and Conjugate Gradient
- Algorithm performance controlled by using learning rate, momentum terms, accuracy and rate of change in the error function
- Use of a test partition to prevent model overfitting
- Network formulas in the output
|
| Standard Scorecards |
Feature Highlights |
| Logistic regression models developed with variables that have undergone weight of evidence transformations can be translated directly into the standard scorecard format |
- Generation of the weight of evidence transformations using decision tree algorithms for binning optimization
- Definition of the scorecard scaling parameters through an intuitive wizard with options including Points to Double the Odds (PDO), base points, reverse scaling and odds at base points
- Scorecards are generated in standard tabular form that can be copied and pasted into Microsoft Office®
- Scorecards can be exported to SQL and SAS code
- Direct scoring on Angoss and external datasets
|
| Market Basket Analysis |
Feature Highlights |
| Market Basket Analysis for discovery and deployment of association rules |
- Deriving association rules from historical data
- Visualizing the degree of attraction or repellence between items
- Charts representing items most strongly associated with a given basket
- Charts visualizing the rank of association rules with respect to their lift or confidence
- Applying association rules to new data to produce recommendations
- Automatic generation of SQL code for association rules, which makes it possible to deploy them in database environments
|
