Incremental Case-Based Reasoning for Classification

Abstract

The focus of this paper is on enhancing the incremental learning of case-based reasoning (CBR) systems. CBR systems can accept new cases and therefore learn as they are being used. If some new attributes are to be added to the available classes, however, the similarity calculations are disturbed and some knowledge engineering tasks should be done to let the system learn the new situation. The attempt here is to make this process automatic and design a CBR system that can accept adding new attributes while it is in use. We start with incremental learning and explain why we need continuous validation of the performance for such dynamic systems. The way weights are defined to accommodate incremental learning and how they are refined and verified is explained. The scheduling algorithm that controls the shift from short-term memory to long-term memory is also discussed in detail.