Extending the Naive Bayes Model Element in PMML: Adding Support for Continuous Input Variables

摘要

The Predictive Model Markup Language (PMML) is the de facto standard to represent data mining and predictive analytic models. With PMML, one can easily share a predictive solution among PMML-compliant applications and systems. PMML as a standard has evolved significantly over the years. PMML 4.1, the language's latest version represents a major leap forward in terms of its ability to represent data post-processing and multiple models. It also provides entirely new model elements for supporting Scorecards and K-Nearest Neighbors. The same is no exception for PMML 4.2, currently being worked on by the Data Mining Group (DMG), the body responsible for maintaining and advancing the PMML standard. PMML 4.2 is bound to offer new elements and increased capabilities. This article describes one of such improvement. In particular, it proposes extending the existing model element for Naive Bayes Classifiers to support continuous input fields. The R Project is a popular choice for data miners to analyze and build predictive models. Naive Bayes is just one of a myriad of model types supported by R. The R e1071 package provides a naiveBayes function to build Naive Bayes Models using categorical as well as continuous fields. The R pmml package has been recently extended to allow for the export of PMML code for objects built with the na?ve Bayes function. For now, it includes a PMML Extension element for continuous fields, but with the release of PMML 4.2, the support will be standardized. This article describes this process in view of our proposal to extend the current model element for Naive Bayes Models.