Artificial Intelligence Research Group

Some Past and Current Projects
Under Development

9. LOKI: Logical Knowledge Inductor
Investigator(s): Mr Ilesh Dattani and Professor Max Bramer

LoKI uses a decomposed model of Principal Component Analysis to pre-process case libraries for significant attribute reduction. The output which forms a subset of the original case library is then used for retrieval. The retrieval algorithms used are derivatives of some of the classic Instance Based Learning algorithms. For inductive retrieval within the model C4.5 is used to build decision trees and rules which can then be applied to the new cases. The model has been rigorously tested on a number of toy datasets and on two datasets within the residential valuation domain. The model further allows the building of weight matrices based on levels of significance on the outcome variable. This is derived by rating the principal components based on probabilistic levels of significance at the 95% level. The results, from empirical work, suggest that the hybrid model provides a better starting point than any ad hoc methods that might be employed. As yet there appears to be very little work in the area of quantifiable measures for the success of Case-Based Reasoning applications underlying frameworks for retrieval. This model is endeavouring to make some progress in this area by, in simplistic terms, using a 'gold standard' by which to measure the retrieved answer.

Dattani, I. Bramer, M.A. (1995) "Case-Based Reasoning: A Technique for Decision Support Systems in Residential Valuation and Construction of Residential Housing" In: Progress in Case Based Reasoning. Ed: Watson, I.D. Springer Verlag Lecture Notes. Berlin pp142-151. Dattani, I, Bramer, M.A. Leonard.G (1995) "A Theoretical Framework for Evaluating the Performance of Case-based reasoning Systems" Proceedings of the 15th British Computer Society Expert Systems Conference 1995. pp 133-145.

Next