Bayesian knowledge extractor

ESPRIT 4 project 29105, Fourth Framework Programme

The aim of this project is to develop an environment able to extract Bayesian Belief Networks (BBNs) from databases. BBNs are one of the most successful formalism for knowledge representation and reasoning and they have been applied to a variety of problems and domains. A BBN provides a graphical representation of decision problems, grounded in probability theory, and they are able to perform prediction, explanation, classification, and decision making. The statistical roots of BBNs give the project an easy access to sound statistical methods for learning. In this way, the methodology underlying the project will blend together well-established statistical theories with the most advanced techniques for machine learning and automated reasoning under uncertainty.

Information Society uses information and generates data. These data are stored in large and fast-growing databases and represent the challenge of effectively navigating this information and the wealth deriving from the exploitation of these data to enhance planning, prediction, and decision making. The newborn field of research known as Knowledge Discovery in Databases (KDD) is meant to meet these challenges. Compared to traditional Data Mining research, KDD aims at developing methods and techniques able to extract reusable knowledge from databases, rather than just to use the database to perform particular tasks.

Given a database, the program will be able to automatically generate a BBN that provides an intuitive graphical representation of the knowledge embedded in the database. The extraction process can be guided and controlled by the user, and prior domain knowledge can be incorporated. Once generated, the BBNs can be deployed as a self-contained intelligent system, able to predict future trends, find explanations for events, and provide normative decisions. The project will develop an integrated environment for the whole BBNs process, including data mining, visualization, knowledge discovery, and deployment of stand-alone intelligent systems. The graphical structure and the symbolic nature of BBNs provides a natural representation of the dependencies identified in the database and allows the user to easily navigate the conceptual relationships discovered in data.

This project brings together world-class research groups and real pioneers of both BBNs and KDD, around the world-leading company in BBNs, to work toward the development of this new tool.

Duration: 18 months
Start date: 1998-12-28
End date: 2000-06-27
Project cost: 0.26 million euro

Coordinator: HUGIN EXPERT A/S

Participants:
HUGIN EXPERT A/S DENMARK
Open University UNITED KINGDOM
Aalborg Universitiet DENMARK
CONSORZIO DI BIOINGEGNERIA E INFORMATICA MEDICA ITALY