PMML

PMML stands for Predictive Model Markup Language. It is an XML-based standard for representing predictive models used in data mining and analytics. PMML enables portable exchange of both data transformations and modeling methods between software tools, supporting deployment in production without bespoke integration code. The standard is maintained by the Data Mining Group and is widely adopted to promote interoperability across platforms.

A PMML document describes a data dictionary using DataField elements, a MiningSchema that maps inputs to model features, and a Model element specific to the algorithm, such as RegressionModel, TreeModel, NeuralNetwork, or ClusteringModel. It also supports transformations through LocalTransformations and GlobalTransformations and enables ensembles via MiningModel with segmentation.

Because PMML files can be generated by model-building tools and consumed by scoring engines that support PMML, they facilitate governance, versioning, and reproducibility across environments. PMML is language- and platform-agnostic, enabling models created in one tool to be deployed in another, from cloud services to on-premises systems.

History and scope: PMML originated in the late 1990s to address model portability. It has evolved through multiple versions, expanding support for algorithms, transformations, and ensemble methods, and remains a common standard for model interchange in analytics ecosystems.

Adoption and tools: PMML is supported by many analytics platforms and libraries, including R, Python ecosystems, KNIME, SAS, IBM SPSS Modeler, and other data-mining tools.