Model-driven data mining engineering: from solution-driven implementations to 'composable' conceptual data mining models

Data mining lacks a general modelling architecture allowing analysts to consider and interpret it as a truly software-engineering process, which would be beneficial for a wide spectrum of modern application scenarios. Bearing this in mind, in this paper, we propose an innovative model-driven engineering approach of data mining whose main goal consists in overcoming well-recognised limitations of actual approaches. The cornerstone of our proposal relies on the definition of a set of suitable model transformations which are able to automatically generate both the data under analysis, which are deployed via well-consolidated data warehousing technology and the analysis models for the target data mining tasks, which are tailored to a specific data-mining/analysis platform. These modelling tasks are now entrusted to the model-transformation scaffolds and rely on top of a well-defined reference architecture. The feasibility of our approach is finally demonstrated and validated by means of a comprehensive set of case studies.