Discovery and Use of Causal Patterns in Databases
David Bell, F.J. McErlean, J. Guan, · Journal of Intelligent Systems · 2000
We consider here a variety of ways of using the attribute values of a database to obtain a single graphical model of cause and effect relationships, from the huge number of possible graphs available (quadratic in the number of attributes, at best).In our method a number of models are obtained from a variety of algorithms.In the paper we also outline our method of reasoning using a generalization of the Bayesian approach to the decomposition of complex problems.Matrices and their products are used to express the reasoning algorithms, making them simple and efficient to implement.These generation and manipulation techniques should be considered as part of an experimental system called Mining Kernel System (MKS) developed by the authors and colleagues, which links such techniques to a variety of data management systems, such as Ingres and Oracle.The primary objective of this paper is to identify the challenges of this novel research, which links knowledge and database engineering techniques, and its associated application pull, and to stimulate further research into the issues arising.