A formal ontology for data mining: Principles, design, and evolution

Data mining (DM) and decision support system (DS) are two relatively independent domains broadly applied to scientific research and business practice. Successful integration of technologies associated with both domains could an intelligent data mining assistant system, which is able to provide intelligent assistance beyond the numbers of DM methods and tools, is an essential step toward a better integration of DM and DS. However, the development of such a system is currently facing two major challenges: the support of non-expert data miner and the definition of DM knowledge. Formalized and computerized ontologies, as a new research area for knowledge conceptualization, possess a great potential to help resolve the above problems. Due to its powerful knowledge representation formalism and associated maintenance mechanism, integrating an ontology into a data mining assistant system will be an effective way of making the system more intelligent and helpful for decision makers.; The objective of the research is to develop an ontology-based approach for data mining. It includes a data mining ontology, which creates a complete data mining domain knowledge base, and an ontology evolution tool, which provides a mechanism to support the ontology development and updating. This research provides a fundamental part of a larger project which aims to develop an intelligent data mining assistant system.; Based on Protege (Stanford University) and the OWL language, a finely designed DM ontology is successfully established. The role of the DM ontology is to represent the data mining knowledge required in the system. Two types of knowledge are represented: data mining domain knowledge that consists of both the methodology and the detailed applicable knowledge of the entire data mining process, and system generated knowledge that consists of data annotation and CBR case representation. To provide more intelligent support for data mining activities, our DM ontology is further integrated with the other two system components: a data warehouse and a case-based reasoning system.; Furthermore, a new ontology evolution methodology is proposed and implemented as a Protege plug-in. This methodology is based on the evolution tasks and the consequence of the change operations. The different change operations and evolution tasks are finely defined in the methodology. The plug-in groups and arranges the necessary steps of most commonly used evolution tasks. It can be used as a step-by-step wizard to guide decision makers to execute ontology-updating tasks.; The results of this research have led to the construction of a fundamental framework for our data mining assistant system and pave the way for a better integration of data mining and decision support system. Furthermore our versatile DM ontology evolution methodology will greatly improve the accuracy, consistency, and efficiency of the evolution tasks. More importantly, this evolution methodology possesses a great potential for further development.