Data Mining in Case-Based Reasoning

Data Mining can be used for a variety of purposes in Case-Based Reasoning. Some uses are sketched here, with increasing complexity. Here we assume that there exists a casebase and in some scenarios also an external database from which further information could be mined.

• Find features for a case (from casebase) - It might be valuable to classify the cases in the casebase for use. This might speed up the computation of similarity metrics in the retrieve step of the CBR cycle, if cases within one class are believed to be more similar than to cases outside of the class.
• Find features for a case (from a database) - A database can be searched to supplement the information given in a case. For instance, Gibbs sampling can be used to fill in missing features in case-data.
• Find domain knowledge (from a database or a casebase) - Domain knowledge might be mined from the data in the form of functions, rules or causal graphs which can later be used by the Case-Based Reasoner when identifying features and explaining away unimportant features in the retrieve step, adapting in the reuse step, or explaining cases in the retain step.
• Construct ``artificial cases'' - one can imagine that it should be possible to construct cases from a database that is not present in a casebase. This would require a tight integration where the DM algorithm would search for patterns in the form of cases, which could be evaluated by a novelty function which gives high values to cases not present in the case- base. These new cases could be used as an automatic brainstorm on new problem situations, to provoke discussions on if they are likely to happen, and on how they can be solved.