Case-Based Reasoning in Data Mining

Two approaches for using CBR in DM are sketched, the first is to use the CBR environment as the environment for the KDD process for the data mining algorithm. The second is to provide the DM with information which is required in order to produce good results.  

• DM search is the case - Data Mining is only one part of the KDD process which can involve accessing several files, cleaning data, and interpreting results. The data mining search may also be time-consuming. The information about the search results and the whole KDD process might be stored in a case so that extra time will not be spent on mining the same information more than once. A need for this approach is signaled after discussions on the CRISP-DM project to develop a standard process model for data mining, which states: ``a standard methodology for data mining must provide a framework for capturing and re-using experiences, and for guiding data mining projects at different levels of skills''.
• CBR provides info - CBR can be used in providing some background knowledge about features in a database, for instance, the weight of features for a classifier can be learnt from the CBR tool. In a Bayesian network, the structure of the network might be set up by the CBR tool (model construction), using its ``expert knowledge'' and the parameters learned using DM algorithms. CBR can also be used to provide utility, validity and novelty functions for the DM algorithm from the domain that the CBR tool is working in (model evaluation).