iCaseViz: Learning case similarities through interaction with a case base visualizer

Abstract

Since the principal assumption in case-based reasoning (CBR) is that "similar problems have similar solutions", learning a suitable similarity measure is an important aspect in CBR. However, learning case-case similarities is often a non-trivial task and involves significant amount of domain expertise. Most techniques that arrive at a pertinent similarity measure are often incomprehensible to the domain experts. These techniques also rarely enable the user to provide expert feedback which can then be utilized to develop better similarity measures. Our work attempts to bridge this knowledge gap by developing an iterative and interactive visualization framework called iCaseViz which learns the domain experts' notion of similarity by utilizing the user feedback. This work is different from similar work in other communities in the sense that it is tailored to cater to the needs of a system built primarily based on the CBR hypothesis. The case base visualizer demonstrated in this paper is also very efficient as it has insignificant delay during real-time user interaction on large case bases. We provide preliminary results on the efficiency of the visualizer and the effectiveness of our similarity learning algorithm on UCI datasets and a real world high dimensional case base. © 2013 Springer-Verlag.