A new strategy for case-based reasoning retrieval using classification based on association

Aljuboori, AS and Meziane, F ORCID: https://orcid.org/0000-0001-9811-6914 2017, A new strategy for case-based reasoning retrieval using classification based on association , PhD thesis, University of Salford.

Download (3MB) | Preview


Cased Based Reasoning (CBR) is an important area of research in the field of Artificial Intelli-gence. It aims to solve new problems by adapting solutions, that were used to solve previous similar ones. Among the four typical phases - retrieval, reuse, revise and retain, retrieval is a key phase in CBR approach, as the retrieval of wrong cases can lead to wrong decisions. To ac-complish the retrieval process, a CBR system exploits Similarity-Based Retrieval (SBR). How-ever, SBR tends to depend strongly on similarity knowledge, ignoring other forms of knowledge, that can further improve retrieval performance. The aim of this study is to integrate class association rules (CARs) as a special case of associa-tion rules (ARs), to discover a set (of rules) that can form an accurate classifier in a database. It is an efficient method when used to build a classifier, where the target is pre-determined.

The proposition for this research is to answer the question of whether CARs can be integrated into a CBR system. A new strategy is proposed that suggests and uses mining class association rules from previous cases, which could strengthen similarity based retrieval (SBR). The propo-sition question can be answered by adapting the pattern of CARs, to be compared with the end of the Retrieval phase. Previous experiments and their results to date, show a link between CARs and CBR cases. This link has been developed to achieve the aim and objectives. A novel strategy, Case-Based Reasoning using Association Rules (CBRAR) is proposed to improve the performance of the SBR and to disambiguate wrongly retrieved cases in CBR. CBRAR uses CARs to generate an optimum frequent pattern tree (FP-tree) which holds a val-ue of each node. The possible advantage offered is that more efficient results can be gained, when SBR returns uncertain answers.

In addition, CBRAR has been evaluated using two sources of CBR frameworks - Jcolibri and Free CBR. With the experimental evaluation on real datasets indicating that the proposed CBRAR is a better approach when compared to CBR systems, offering higher accuracy and lower error rate.

Item Type: Thesis (PhD)
Schools: Schools > School of Computing, Science and Engineering
Funders: University of Baghdad
Depositing User: AS Aljuboori
Date Deposited: 15 Jan 2018 11:37
Last Modified: 27 Aug 2021 20:39
URI: http://usir.salford.ac.uk/id/eprint/42230

Actions (login required)

Edit record (repository staff only) Edit record (repository staff only)


Downloads per month over past year