| Enhanced Descriptive Statistics | | |
| | Summary statistics: full, brief, stratified | • | • |
| | Charts and histograms | • | • |
| Improved User Interface | | |
| | New setup activity window | • | • |
| | Ability to control default settings | • | • |
| Model Building | | |
| | Additional fraction for auto validation | • | • |
| | Select/reject predictors directly from variable importance list | • | • |
| | Lag variables and autoregression | • | • |
| Cross Validation | | |
| | User-controlled cross validation bins | • | • |
| | Repeated CV using different random seeds (BATTERY CVR) | • | • |
| Missing Value Analysis | | |
| | Automatically add missing value indicators | • | • |
| | Allow “missing” as a legal discrete level | • | • |
| Model Evaluation & Reporting | | |
| | Monte Carlo testing (BATTERY MCT) | • | • |
| | ROC curves and variance of ROC measure | • | • |
| | Display learn, test, or pooled results | • | • |
| | Gains chart: show perfect model curve | • | • |
| Scoring and Translation | | |
| | New model translation formats: Java and PMML | • | • |
| Automated Model Search: BATTERY | | |
| | BASIS: Varies the number of basis functions | • | • |
| | CV: Varies the number of folds used for cross-validation | • | • |
| | CVR: Repeats cross validation using different random number seeds | • | • |
| | DRAW: Randomly draws the learn sample from the “main” learn sample | • | • |
| | FLIP: Reverses the learn and test samples | • | • |
| | INTER: Varies the number of interactions | • | • |
| | MCT: Monte Carlo shuffling of the target variable | • | • |
| | MINSPAN: Varies the minimum span | • | • |
| | SAMPLE: Progressively reduces the size of the learn sample | • | • |
| | SPEED: Varies the MARS speed parameter through all meaningful values | • | • |
| | TARGET: Models each variable as a target, using all other variables as predictors | • | • |
| | KEEP: Randomly selects a subset of variables as predictors for each model | | • |
| | LOVO: Leaves one predictor out of the model each time | | • |
| | ONEOFF: Models the target as a function of one predictor at a time | | • |
| | SHAVING: Sequentially removes predictors from the model based on their importance | | • |
