Evaluation, Revision, and Learning

The first part of this chapter is basic and is of interest to readers who want to build or maintain a CBR system. The second part provides background knowledge. It deals with revising the methods if something is definitely wrong and with improving CBR systems by machine learning if the results are weak. In this chapter methods for evaluating, revising and improving CBR are discussed in that order. Evaluation detects the weaknesses of a system. Revise is an activity from outside for removing individual faults and related problems. Learning considers improvement of the whole system. It deals all knowledge containers of a CBR system as well as the overall system. For improving the case base, the three classic algorithms IB1, IB2, and IB3 are presented. The learning of similarities is concerned with similarity relations, weights, and local similarities. As a crucial step in similarity learning, the learning of weights is discussed. The chapter also presents some machine learning methods that are used within and in integration with CBR, including regression learning, artificial neural networks, genetic algorithms, clustering algorithms, and Bayesian learning. These methods that are used within and in integration with CBR.