Variable precision rough set theory decision support system: With an application to bank rating prediction

Griffiths, Benjamin 2008. Variable precision rough set theory decision support system: With an application to bank rating prediction. PhD Thesis, Cardiff University.

PDF - Accepted Post-Print Version Download (26MB)

Abstract

This dissertation considers, the Variable Precision Rough Sets (VPRS) model, and its development within a comprehensive software package (decision support system), incorporating methods of re sampling and classifier aggregation. The concept of /-reduct aggregation is introduced, as a novel approach to classifier aggregation within the VPRS framework. The software is applied to the credit rating prediction problem, in particularly, a full exposition of the prediction and classification of Fitch's Individual Bank Strength Ratings (FIBRs), to a number of banks from around the world is presented. The ethos of the developed software was to rely heavily on a simple 'point and click' interface, designed to make a VPRS analysis accessible to an analyst, who is not necessarily an expert in the field of VPRS or decision rule based systems. The development of the software has also benefited from consultations with managers from one of Europe's leading hedge funds, who gave valuable insight, advice and recommendations on what they considered as pertinent issues with regards to data mining, and what they would like to see from a modern data mining system. The elements within the developed software reflect each stage of the knowledge discovery process, namely, pre-processing, feature selection, data mining, interpretation and evaluation. The developed software encompasses three software packages, a pre-processing package incorporating some of the latest pre-processing and feature selection methods a VPRS data mining package, based on a novel "vein graph" interface, which presents the analyst with selectable /-reducts over the domain of / and a third more advanced VPRS data mining package, which essentially automates the vein graph interface for incorporation into a re-sampling environment, and also implements the introduced aggregated /-reduct, developed to optimise and stabilise the predictive accuracy of a set of decision rules induced from the aggregated /-reduct.

Item Type:	Thesis (PhD)
Status:	Unpublished
Schools:	Business (Including Economics)
Subjects:	H Social Sciences > HJ Public Finance
ISBN:	9781303224522
Date of First Compliant Deposit:	30 March 2016
Last Modified:	12 Feb 2016 23:13
URI:	https://orca.cardiff.ac.uk/id/eprint/55175

Actions (repository staff only)

Edit Item