Using the Analysis Studio regression wizard allows fast production and on the fly model comparison in data mining projects. The software produces 5 different regression models (linear, polynomial, logarithmic, exponential and power) allowing you to pick the best model of choice on the fly. At the process of producing the regression model, you may go back, change the selected variable(s) and produce better models in less time.
As a final step, after variables are picked and the model is chosen publish the model.
Once a model is published you may scrutinize model stability, test model performance in "What if?" scenarios and produce sensitivity tables.Reduce model generation time: During the procedure of model generation, you may go back and forth and change selected settings from each step.
Step 1 Pick variables – Pick the explained variable and select the column(s) that explain it. In this step you can select the variable insertion method (all selected variables, stepwise selection by p-value or stepwise selection).
Step 2 Compare models – Preview the dependent explained variable and selected independent variables as they are expressed in different regression models. Here you may select the final model that will be used.
Step 3 Preview – Preview the selected model parameters as well as variable specific parameters.
- Regression method
- Variable selection method
- Multiple R
- Adjusted R-square
- Std err
- Durbin - Watson
- Standard Error
- Lower limit
- Upper limit
Step 4 Publish – Publish the model as a part of your Analysis studio data mining project. Once a model is published you may scrutinize and verify the model, comprising independent variables and model behavior under different values.
What if ? - The "What If?" scenario screen allows you to test the model with different values of different variables. This screen lets you intuitively understand the impact of variable changes on model outcome.
Sensitivity Table – Compute a range of values for a selected pivot variable and view the model outcome for the entire range.
Anomaly - Find the data that poorly responds to the regression line (i.e. data that does not "behave" as expected by the regression model). This feature is used for model refining and pre model filtering.