Case Base Size and Overall Competence: Incremental Increase and Similarity Threshold Selection on a Data Set

This paper builds on prior case-based reasoning (CBR) research into the effect of incremental increase of case library size on overall system competence. We explore the effect of gradually increasing the size of the case base on the total accuracy of a CBR algorithm, using a large numeric data set as an example. We use a standard mathematical definition of system competence in our study of the data set and compare these results to those obtained in prior research on several other sets. We also use a strategy for adjusting the size of the similarity threshold, or maximum distance between a test case and a training point for which the match between the two can be considered successful. This makes use of both data normalization and consideration of maximum possible error. As expected, increasing the size of the case library had the effect of increasing competence and accuracy, but this technique did not always produce the most intuitive results.