Adaptive case-based reasoning using retention and forgetting strategies
Case-based reasoning systems need to maintain their case base in order to avoid performance degradation. Degradation mainly results from memory swamping or exposure to harmful experiences and so, it becomes vital to keep a compact, competent case base. This paper proposes an adaptive case-based reasoning model that develops the case base during the reasoning cycle by adding and removing cases. The rationale behind this approach is that a case base should develop over time in the same way that a human being evolves her overall knowledge: by incorporating new useful experiences and forgetting invaluable ones. Accordingly, our adaptive case-based reasoning model evolves the case base by using a measure of “case goodness” in different retention and forgetting strategies. This paper presents empirical studies of how the combination of this new goodness measure and our adaptive model improves three different performance measures: classification accuracy, efficiency and case base size.