We exemplify in this paper, how a discovery system is applied to the analysis of simulation experiments in practical political planning, and show what kind of new knowledge can be discovered in an application area that differs from others by the high amount of knowledge that the analyst holds already about the process that generates the data. Subgoals like "low classification accuracy", "high homogeneity", "disjoint rules", etc. are introduced into Explora, to select between different statistical tests for each pattern and several search algorithms, allowing the user to adapt the discovery process to the special requirements of the application.
The combination of discovery with simulation is endowed with the main characteristics of both Knowledge Discovery in Databases (KDD) and Automated Scientific Discovery (ASD), i.e. discovery in large databases and experimentation. Analysing a real system with simulation models allows to freely set the experimental conditions. In distinction to the usual KDD assumption of fixed given data, when combining discovery with simulation, additional data can be generated by running new simulations according to the needs of the discovery component.
First, we identify four tasks relevant for exploring simulation experiments which can be supported by discovery methods. Then we describe the application of socio-economic modeling for political planning. Third, we demonstrate for a simple law (financial support of families with children), how the current state of Explora is used for this application. Finally, we discuss some new approaches of Explora to deal with subgoals and outline further work.
[1]
Willi Klösgen,et al.
Sozio-ökonomische Simulation und Planung. Entwicklungsstand und Computerunterstützung
,
1985,
Inform. Spektrum.
[2]
P. Langley,et al.
Computational Models of Scientific Discovery and Theory Formation
,
1990
.
[3]
William Frawley,et al.
Knowledge Discovery in Databases
,
1991
.
[4]
Willi Klösgen,et al.
Problems for knowledge discovery in databases and their treatment in the statistics interpreter explora
,
1992,
Int. J. Intell. Syst..
[5]
Leo Breiman,et al.
Classification and Regression Trees
,
1984
.
[6]
Gregory Piatetsky-Shapiro,et al.
Discovery, Analysis, and Presentation of Strong Rules
,
1991,
Knowledge Discovery in Databases.
[7]
F. Gebhardt.
Discovering interesting statements from a database
,
1994
.
[8]
W. Scott Spangler,et al.
Learning Useful Rules from Inconclusive Data
,
1991,
Knowledge Discovery in Databases.
[9]
Friedrich Gebhardt,et al.
Choosing among competing generalizations
,
1989
.
[10]
Joachim Merz,et al.
Microanalytic simulation models to support social and financial policy
,
1986
.