An Extended Predictive Model Markup Language for Data Mining

摘要

Common data mining metadata benefits sharing, exchanging and integration among data mining applications. The Predictive Model Markup Language PMML facilitates the exchange of models among data mining applications and becomes a standard of data mining metadata. However, the evolution of models and extension of products, PMML needs large number of language elements and leads to conflicts in PMML based data mining metadata inevitably. This paper presents an extended predictive model markup language EPMML for data mining, which is designed to reduce the complexity of PMML language elements. The description logic for predictive model markup language DL4PMML that belongs to the description logic family, is the formal logical foundation of EPMML and makes it possess strong semantic expression ability. We analyze how EPMML describe data mining contents in detail. Some experiments expatiate how EPMML based data mining metadata support automatically reasoning and detect inherent semantic conflicts.