Inductive Databases: Towards a New Generation of Databases for Knowledge Discovery

Data mining applications are typically used in the decision making process. The Knowledge Discovery Process (KDD process for short) is a typical iterative process, in which not only the raw data can be mined several times, but also the mined patterns might constitute the starting point for further mining on them. These are the premises that lead Imielinski and Mannila in "A database perspective on knowledge discovery", CACM, 39(11), 1996, to propose the idea of inductive database, a general-purpose database in which both the data and the patterns can be represented, retrieved and manipulated. The goal of inductive databases is to assist the deployment of the KDD process and integrate several heterogeneous data mining and data analysis tools. In this paper we overview the current state of the art of the research in databases support for KDD. We mean database standards for KDD, APIs for data mining, ad-hoc query languages and constraint-based query optimization. Our look is essentially from an academic point of view but also from an industrial one.