GrobianSchool of Information and Software Engineering, University of Ulster, Newton-abbey, IRELANDFB Psychologie/Methodenlehre, Universitat Osnabruck, Osnabruck, GERMANY
Abstract
GROBIAN is an implementation of the rough information analysis ROUGHAIN described in [2]. The acronym GROBIAN comes from the German “GROBmengen Informations ANalyse”, which is the some as “ROUGH set Information ANalysis - and the meaning of the German GROBIAN and the English ROUGHIAN is (roughly) the same. GROBIAN is bilingual (English/German). Beside the standard procedures of Rough Set Data Analysis, such as reduct analysis gamma and alpha - statistics and rule generation, GROBIAN has the following enhanced features:
- A randomization test for rule significance including a sequential testing procedure,
- A simple method for data filtering which may increase the significance of rules,
- Model selection based on information theoretic entropy,
- Jack-knife vlidation,
- Training/Testing validation.
A menu driven recording and restriction procedure for data manipulation is an additional feature of GROBIAN.
Developers
- Main developer: Günther Gediga, FB Psychologie and Institut für semantische Informationsverarbeitung, Universität Osnabrück, ggediga@Luce.psycho.uni-osnabrueck.de
- Ivo Düntsch, Faculty of Informatics, University of Ulster, I.Duentsch@ulst.ac.uk
History
- V 0.01-0.06 Versions for internal use only,
- V 0.07 First published GROBIAN version, including significance testing, rough filtering, rough entropy (July 1996),
- V 0.08-0.13 Several minor changes and bug removal,
- V 0.14 Removed the RSL-library from most parts of GROBIAN,
- V 0.15 Added JACK-KNIFE validation procedure,
- V 0.16 Added batch processing from a list reducts,
- V 0.17 Sequential Randomization Test as now available for significance testing. Added TRAINING SET-TESTING SET validation procedure,
- V 0.18 Result windows can handled as objects (Oct 1997).
Input
One table which may be portioned by data restrictions to enable many tables-applications. ASCII-files, RSL-SYS-files and specific GROBIAN files are possible input formats. There is no direct interaction with DBMS system (except via ASCII export-import).
Output
The result of the chosen procedure on screen and as an ASCII file.
System requirements
- Processor: Intel 80386 or later,
- MS Windows family (3.1, Win95, NT),
- >= 12 mb RAM,
- >= 10 mb disk space,
Evaluation criteria applied by the system
- Approximation quality gamma,
- Statistical significance of gamma,
- Drop of gamma,
- Statistical significance of drop gamma,
- Jack-knifed predicted success (lower bound and expectation),
- Training-testing set prediction success (lower bound and expectation)
- Information analysis (standard and rough information),
- Statistical significance of information measures.
Differences of other systems
Emphasis on non-invasive data filtering and statistical validation of e-reducts of information system. |