Applying multiple classification ripple round rules to a complex configuration task

  • Authors:
  • Ivan Bindoff;Byeong Ho Kang

  • Affiliations:
  • School of Computing and Information Systems &, School of Pharmacy, University of Tasmania, Australia;School of Computing and Information Systems, University of Tasmania, Australia

  • Venue:
  • AI'11 Proceedings of the 24th international conference on Advances in Artificial Intelligence
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

A new expert systems methodology was developed, building on existing work on the Ripple Down Rules (RDR) method. RDR methods offer a solution to the maintenance problem which has otherwise plagued traditional rule-based expert systems. However, they are, in their classic form, unable to support rules which use existing classifications in their rule conditions. The new method outlined in this paper is suited to multiple classification tasks, and maintains all the significant advantages of previous RDR offerings, while also allowing the creation of rules which use classifications in their conditions. It improves on previous offerings in this field by having fewer restrictions regarding where and how these rules may be used. The method has undergone initial testing on a complex configuration task, which would be practically unsolvable with traditional multiple classification RDR methods, and has performed well, reaching an accuracy in the 90th percentile after being trained with 1073 rules over the course of classifying 1000 cases, taking ˜12 expert hours.