Maintaining case-based reasoning systems using a genetic algorithms approach

https://doi.org/10.1016/S0957-4174(01)00035-5Get rights and content

Abstract

The success of a case-based reasoning (CBR) system largely depends on an effective maintenance of its case-base. This study proposes a genetic algorithms (GAs) approach to the maintenance of CBR systems. This approach automatically determines the representation of cases and indexes relevant attributes to grasp the rapidly changing environment around the system. In this study, the proposed model is applied to stock market analysis. Experimental results show that the proposed model outperforms conventional CBR systems.

Introduction

Case-based reasoning (CBR) is a reasoning technique that reuses past cases to find a solution to the new problem. Previous studies suggested that CBR is preferred method for ill-structured managerial decisions (Deng, 1996). This suggestion stems from the fact that historical cases are a manifestation of individual or organizational wisdom. The nature of historical cases, however, evolves by the change of domain knowledge and environments. In practice, domain knowledge and environments change, making the previous knowledge useless (Will, McQuaig & Hardaway, 1994). Weitz and DeMeyer (1990) suggested that an evolutionary process is needed to get important information and to refine the knowledge-base in expert systems. Case-base must also be refined and be revised for this reason.

Case-base maintenance (CBM) is the process of refining the case-base of a CBR system to enhance the performance and integrity of the system (Leake & Wilson, 1998). The case-base may need to reflect the change of domain knowledge and environments by proper CBM policies.

This study proposes a genetic algorithms (GAs) approach to case representation and indexing for CBM. This approach automatically determines proper case representation and selects relevant indexes for the evolving environments. The GA globally searches optimal or near-optimal representation and indexes. In this paper, the proposed approach is applied to stock market analysis, then experimental results are presented.

The rest of the paper is organized as follows. Section 2 reviews prior research on maintaining and updating CBR systems. Section 3 proposes a GA approach to case representation and indexing for CBM. In this section, the benefits of the proposed approach are described. Section 4 describes the application of the proposed approach. In this section, the empirical results are summarized and comparative analysis of the proposed and a conventional approach is described. In the final section, the conclusions and limitations of this study are presented.

Section snippets

Prior research on case-base maintenance

Many researchers have addressed the issue of CBM. Aha and Breslow (1997) proposed an automated inductive approach that revises conversational case libraries. They used a top-down algorithm to induce a decision tree from the cases. Smyth (1998) classified the method of maintenance in CBR as efficiency-directed and competence-directed maintenance. He proposed the approach to maintenance which is based on the deletion of harmful and redundant cases from the case-base. He suggested that the

A case-base maintenance model with the GA

CBR is composed of the steps of the representation, indexing, retrieval, and adaptation of cases. Each step contributes the maintaining process of CBR systems. This study proposes a new CBM policy for the step of case representation and indexing based on the GA.

Application to stock market data

This section applies the CBM model with the GA to stock market prediction. Many studies on stock market prediction using artificial intelligence techniques have been performed in the past decade. The noise and non-stationary characteristics in stock market data are one of major reasons for the poor performance by some previous studies. If these factors are not appropriately controlled, the prediction system does not produce a good performance. When the prediction is executed using long-term

Concluding remarks

As mentioned earlier, case-base may grasp the change of organizational knowledge and environments through proper CBM policies. Prior studies have addressed the issue of CBM. Their studies, however, mainly relied on the deletion and revision of irrelevant and redundant cases. This study proposes the GA approach to CBM policy. This approach considers the relationship between independent and dependent attribute. In addition, this approach does not eliminate specific cases but revises the

Acknowledgements

The authors would like to thank the Korea Science and Engineering Foundation for supporting this work under Grant No. 98-0102-08-01-3.

References (24)

  • R.J. Bauer

    Genetic algorithms and investment strategies

    (1994)
  • C.E. Brown et al.

    Applying case-based reasoning to the accounting domain

    International Journal of Intelligent Systems in Accounting, Finance and Management

    (1994)
  • Cited by (49)

    • Evaluating Case-Base Maintenance algorithms

      2014, Knowledge-Based Systems
      Citation Excerpt :

      Below, we shall comment on the most common evaluation method on the CBR field. In [4,27] the accuracy of the maintained CBR system is studied using Hold-Out, and a k-nearest neighbour (K-NN) is used as a classifier. Following the same evaluation method, in [32] compare the retrieval time and the size reduction rate.

    • Knowledge-based modeling for disruption management in urban distribution

      2012, Expert Systems with Applications
      Citation Excerpt :

      Therefore, based on the characteristics of different knowledge included in the modeling process, different knowledge representation schemes will be adopted in this research. With respect to knowledge-based modeling, except for case-based modeling (Chang, Liu, & Lai, 2008; Kim & Han, 2001), ontology-based modeling (Bobillo, Delgado, & Gómez-Romero, 2008), etc., which are suitable for other problems, algebraic modeling language is the branch suitable for optimization problems. Algebraic modeling language was pioneered by AMPL (Fourer, Gay, & Kernighan, 1990), GAMS (Brooke, Kendrick, & Meeraus, 1992), AIMMS (Bisschop, 2006; Bisschop & Entriken, 1993), and followed by some complementary methods, like AMPL extension language (Desrochers, Hones, Lenstra, Savelsbergh, & Stougie, 1999), MILANO (Sengupta, 2009), etc.

    View all citing articles on Scopus
    View full text