Knowledge Discovery from Databases

As databases are pervading every parcel of reality, they record what happens in and around organizations all over the world. Databases store the detailed history of organizations, institutions, governments and individuals. An efficient and agile analysis of the data recorded in a database can no longer be done manually. Knowledge Discovery from Databases (KDD) is a collection of technologies which aim at extracting nontrivial, implicit, previously unknown, and potentially useful information (Fayyad, Piatetsky-Shapiro, Smyth & Uthurusamy, 1996) from raw data stored in databases. The extracted patterns, models or trends can be used to better understand the data, and hence the context of an organization, and to predict future behaviors in this context that could improve decision making. KDD can be used to answer questions such as: Is there a group of customers buying a special kind of products? Which sequence of financial products improves the chance of contracting a mortgage? Which telephone call patterns suggest a future churn? Are there relevant associations between risk factors in coronary diseases? How can I assess my e-mail messages more or less likely to be spam (junk mail)? The previous questions cannot be answered by other tools usually associated to database technology such as OLAP tools, decision support systems, executive information systems, etc. The key difference is that KDD does not convert information into (more aggregated or interwoven) information, but generates inductive models (under the form of rules, equations or other kind of knowledge) that could be sufficiently consistent with the data. In other words, KDD is not a deductive process but an inductive one.