Integrating the development of data mining and data warehouses via model-driven engineering

Data mining is one of the most important analysis techniques to automatically extract knowledge from large amount of data. Nowadays, data mining is based on low-level specifications of the employed techniques typically bounded to a specific analysis platform. Therefore, data mining lacks a modelling architecture that allows analysts to consider it as a truly software-engineering process. Bearing in mind this situation, we propose a model-driven approach which is based on (i) a conceptual modelling framework for data mining, and (ii) a set of model transformations to automatically generate both the data under analysis (that is deployed via data-warehousing technology) and the analysis models for data mining (tailored to a specific platform). Thus, analysts can concentrate on understanding the analysis problem via conceptual data-mining models instead of wasting efforts on low-level programming tasks related to the underlying-platform technical details. These time consuming tasks are now entrusted to the model-transformations scaffolding. The feasibility of our approach is shown by means of a hypothetical data-mining scenario where a time series analysis is required.

This work has been supported by the ESPIA (TIN2007-67078) project from the Spanish Ministry of Education and Science, and by the QUASIMODO (PAC08-0157-0668) project from the Castilla-La Mancha Ministry of Education and Science (Spain). Jesús Pardillo and Jose-Norberto Mazón are funded by the Spanish Ministry of Education and Science under FPU grants AP2006-00332 and AP2005-1360, respectively.