Attribute reduction and its performance evaluation in case-based reasoning

It is the attribute sets which describe the various factors that determine the system's character, and it is a critical factor for system performance to select and reduce the attributes in Case-based reasoning (CBR). On the basis of the attribute-oriented reduction techniques, this paper focuses on the two entropy-based attribute-selecting strategies after analyzing the attribute reduction techniques. Using a method combining stratified k-fold cross-validation and k-nearest neighbor (k-NN), five schemas were designed to evaluate the performance of the above two attribute selecting strategies from different angles. Experimental results indicate that the entropy-based attribute selection strategy can find an attribute subset which can separate the case classes sufficiently and effectively.