Attribute-Oriented Induction (AOI) is a descriptive data mining technique allowing generalization of original attribute values to discover high-level (abstract) knowledge from information stored in large databases. The hierarchical aggregation of similar data in the AOI approach has a data-driven character, which can be indirectly influenced by experts whose knowledge is reflected in concept hierarchies utilized in the AOI process. In this work we analyze the utilization of fuzzy data structures and fuzzy relations to achieve a more flexible representation of background knowledge reflecting the relationships among the attribute values in the generalized domains. We introduce a formal framework of fuzzy generalization and present desired properties of an Attribute-Oriented Fuzzy Induction (AOFI) method that allows better modeling of real-life dependencies occurring among the generalized data.; We investigate the applicability of an attribute-oriented induction approach for acquisition of generalized knowledge from data stored in fuzzy relational databases. We analyze the proximity-based and similarity-based fuzzy database schemas and use the original properties of those databases to support the AOI. We also establish a new method for generalization of tuples with set-valued data, which represent imprecise information. In our approach we take full advantage of the implicit knowledge about the similarity of originally stored attribute values, included by default in both analyzed fuzzy database schemas.; The approach developed in the first part of this dissertation is demonstrated by practical data mining project. We use the Attribute-Oriented Fuzzy Induction approach to mine concise information at a high level of abstraction from the data stored in the Toxic Release Inventory, a database managed by U.S. Environmental Protection Agency (EPA).
展开▼
