Encapsulating classification in an OODBMS for data mining applications

Classification is an important task in data mining. Encapsulating classification in an object-oriented database system requires additional features: we propose multiobjects and schema evolution. Our approach allows us to store classification functions, and to store instances of each group in order to retrieve them later. Since the database is operational, it allows us also to perform dynamic classification, i.e. add/remove instances to/from groups over time. Moreover it allows us to update classification functions (if we choose another population sample or apply another classifier) and have the instances of groups consequently reclassified. We illustrate our approach with a target mailing application.