Case-Base Maintenance for Concept Drift
- Publication Type:
- Conference Proceeding
- WASET Issue 72: Proceedings of the International Conference on Information Systems, 2012, pp. 333 - 340
- Issue Date:
As the evolving nature of a real data stream, so called concept drift, and its accumulating volume, any deployed Case-Based Reasoning (CBR) system will nead to have procedures for Case-Base Maintenance (CBM). Traditional CBM methods for handling concept drift cannot well distinguish between noise and true concept drift, some of them may even run under the risk of losing case-base competence. Motivated by these two problems, wc present a twostage CBM approach. We propose a Noise-Enhanced Fast Context Switch (NEFCS) algorithm, which targets to remove noise under dynamic environment as Stage 1. We also propose a Recursive Conservative Redundancy Removal (RCRR) algorithm, which removes redundant cases in a recursively unifonn way while keeping the case·base coveragc as Stage 2. Experimental evaluations on realworld datasels show that our approach significant improves the perfonnance compared with other CBM approaches.
Please use this identifier to cite or link to this item: