Predictive model representation and comparison: Towards data and predictive models governance

Abstract

The increasing variety of data mining tools offers a large palette of types and representation formats for predictive models. Managing the models becomes then a big challenge, as well as reusing the models and keeping the consistency of model and data repositories because of the lack of an agreed representation across the models. The flexibility of XML representation makes it easier to provide solutions for Data and Model Governance (DMG) and support data and model exchange. We choose Predictive Toxicology as an application field to demonstrate our approach to represent predictive models linked to data for DMG. We propose an original structure: Predictive Toxicology Markup Language (PTML) offers a representation scheme for predictive toxicology data and models generated by data mining tools. We also show how this representation offers possibilities to compare models by similarity using our Distance Models Comparison technique. This work is ongoing and first encouraging results for calculating PTML distance are reported hereby.