Using NIAR k-d Trees to Improve the Case-Based Reasoning Retrieval Step

摘要

Case retrieval is one important step in the case-based reasoning cycle. Up to now, several algorithms have been proposed for the indexing of cases, since the original indexing approach of k-d trees came up in literature. Main approaches propose the use a precomputed binary search tree to get an average logarithmic time effort in searching. The proposal presented in this paper consists of an indexing algorithm based on the principle of binary search trees for efficient case retrieval according to a given similarity measure called sim. The proposed NIAR k-d tree algorithm embodies two main steps based on the computation of the average value of the corresponding attribute among the subtree cases, and selecting for that attribute, the value of the Nearest Instance/case to the Average as the Root (partition value). Experimental results with some databases have shown that the retrieval step in NIAR k-d tree is faster than the standard k-d tree approach. The time efficiency, the depth and breadth in both trees are analyzed. The results obtained depict a significant difference of levels in the trees. The presented approach is implemented within a current research work on introspective reasoning framework for case-based reasoning in continuous domains.