Case-based reasoning (CBR) is a problem-solving paradigm and is based on the idea that solving new problems can be based on similar past problems. CBR is a pervasive behavior in everyday human problem solving; almost all reasoning is based on past anecdotal experiences. Given a problem, one retrieves from memory more or less identical cases relevant to solving a specific problem. These retrieved cases may consist a problem, a solution, and annotations about how the solution was derived.
To apply CBR into a maintenance environment a rich set of good practices can be use similar to a human memory. To reuse a historical case from a tacit knowledge database (case base) may involve some adaptation of to fit the solution into the new situation. The next step is to test the new solution and, if necessary, revise. If the solution has been successfully adapted, the resulting experience is stored as a new case. Cases can be distinguished with respect to the homogeneity of the cases. Homogenous cases share the same attributes. Heterogeneous case have different attributes but may share some.
At first glance, CBR may seem similar to the algorithms of machine learning. Like algorithm, CBR starts with a set of cases; it forms generalizations by identifying commonalities between a problem and the derived solution. The key difference between CBR and algorithms lies in process step where the generalization is made. Algorithm draws generalizations from a set of training examples before the problem is even known. The difficulty with algorithm is in anticipating the different generalizations directions in the examples. This is in contrast to CBR, which delays the process of generalization after testing. CBR therefore is a good method for maintenance situation containing rich, complex situation in which there are myriad ways to solve a case.
In situations without statistically relevant data for backing and implicit generalization, there is no guarantee that other ways of generalization are correct. However, all reasoning where data is too scarce for statistical relevance solutions are inherently based on anecdotal evidence.