Self-Validating Ensemble Modeling¶
Self-Validating Ensemble Modeling (SVEM) is Predictum's groundbreaking platform for fitting stable predictive models in situations where there is little data. The SVEM platform is packaged as a JMP add-in file.
The SVEM methodology constructs an ensemble model. It fits a collection of models by using the existing data for self-validation; one copy of the data set is used for training and a second copy for validation. Prior to each model fit, random fractional bootstrapped weights are assigned to the observations so that the training and validation data are inversely correlated. The collection of bootstrapped models constructed in this manner is averaged to produce the final model. Being an ensemble model, the final model is robust and has good predictive performance.
For more information about SVEM's methodology and algorithm, see the published research paper. A free, pre-release version of the paper is available on arXiv's website, which is a distribution service for scholarly articles.
The SVEM report enables you to save the final model and an estimate of prediction error. You can explore and find optimal settings for the final model using the Prediction Profiler, which is accessible from the SVEM report.
SVEM Pathways for Model Building¶
The SVEM toolkit supports two pathways for model fitting:
-
Neural Networks pathway: This is based on the Neural modeling technique in JMP software, both the standard and Pro versions. The Neural pathway enables you to fit flexible, non-linear models.
-
Generalized Regression pathway: This is based on the Generalized Regression modeling technique in JMP Pro software only. The Generalized Regression pathway enables you to fit various modeling situations and provides variable selection techniques, including Lasso and Elastic Net, that address highly correlated predictors. You can specify a variety of distributions for your response variables.
System Requirements¶
SVEM is required to be run with JMP software, but the system requirements are different depending on which pathway you choose to use and which version of JMP software you have installed.
| JMP Software Version | JMP Version Number | SVEM Pathways Supported |
|---|---|---|
| JMP Pro | 15, 16, 17, and 18 | Generalized Regression and Neural Networks |
| JMP standard | 15, 16, 17, and 18 | Neural Networks only |
Installation¶
To install the SVEM application, navigate to the directory where you saved the SVEM add-in file and double-click on it. Once installed, SVEM is available from the Predictum menu > Self-Validating Ensemble Modeling.