Knowledge Engineering

Case Based Reasoning (CBR) is more suitable for the following problems:

• The problem domain is complex and not amenable to complete mathematical modeling.



• An explicit model is extremely difficult to elicit and represent with rules.



• Historical data exists within the organization.



• The domain experts have considerable difficulty in writing down the decision rules. But, they are comfortable in providing well-proven heuristics and experiences, to incorporate into the case base as cases. They have little difficulty in recalling concrete cases, which they have encountered in the past.



• The domain is such that it needs reasoning for the system's solutions.

The main similarity between Artificial Neural Networks (ANN) and Case Based Reasoning (CBR) is that both do not need an explicit model. Both CBR and ANN techniques do not have to go through the knowledge-acquisition bottleneck. ANN is essentially a data mining technique, which can work directly from the data.

But, the main criticism against ANN is that it works as a “Black Box”, so they suffer from a lack of transparency. Validity of the systems decision cannot be judged because of the nature of the inner workings, the output of the network is a function of weighted vectors that depends on the network's architecture and the learning mode used. So, it becomes very difficult to use ANN for diagnosis applications, as most of the diagnosis needs an explanation for the result obtained.

ANN are not suitable when background domain knowledge has to be taken into account, whereas in CBR domain knowledge can be incorporated in the form of knowledge-guided clustering. Neural networks cannot cope with complex structures and in order to perform well the coverage of the domain has to be exhaustive during the "learning" phase. CBR does not need an exhaustive coverage of the domain, as cases can be added to the case-base incrementally.

Rule Based Reaoning (RBR) requires us to elicit an explicit model of the domain. As we all know and have experienced, knowledge acquisition has a set of associated problems. In contrast, Case Based Reasoning (CBR) does not require an explicit model. Cases that identify the significant features are gathered and added to the case base during development and after deployment. This is easier than creating an explicit model, as it is possible to develop case bases without passing through the knowledge-acquisition bottleneck.

Domain experts would have accumulated knowledge over the years through experience. It would be a blunder mistake if we do not use them to our advantage. But the difficulty lies in getting the experts to list down the decision rules which they use. It is a Herculean task to comprehensively recall all the tacit rules which they have come to adopt. However, they usually have little difficulty in recalling concrete cases, which they have encountered in practice. Thus, their mental set appears to be oriented towards a Case Based Reasoning approach.

Developing a CBR system is much faster and easier than constructing a rule-based equivalent. Case bases do not have to be complete when they are deployed for use, as even non-computer experts can add cases to the existing structure.

Maintenance with a Rule Based System may be a nightmare. If the rules are not written clearly, it would lead to many sleepless nights of debugging. Maintenance with Case Based Systems are much easier and straightforward.

When rules are added or deleted from a rule-based system, the system has to be checked for conflicting rules and redundant rules. An addition or deletion of a case from the case base does not any further checking or debugging. But it have to be noted that while it does not affect the system’s functioning, it may have an impact on the outcome of the system.

In an earlier post, we saw what exactly is Case-Based Reasoning (CBR). Here, we are going to see how CBR approaches a problem.

CBR revolves around case base, a database of past cases, that typically contains cases with problem description, possible causes and solutions. Don't confuse yourself with case base and database. They are similar, but not the same. Database works on crisp data and exact solution match, whereas CBR is based on fuzzy match. CBR has the ability to learn and improve, whereas databases do not.

In CBR, a new problem is solved by finding a similar past case, and reusing it in the new problem situation. Note the word similar. CBR would give a solution even if you don't get an exact match. If you accept the solution, then it goes into the case base as another case. If the solution is not accepted, it gives CBR a learning step.

The following four steps summarize how a problem is solved in CBR:

Case-Based Reasoning (CBR) is a problem solving model that is fundamentally different from other major Artificial Intelligence (AI) techniques.

CBR does not rely solely on the general knowledge of a problem domain. CBR is able to make use of the specific knowledge of previously experienced concrete problem situations. These previous situations are referred to as cases in CBR.

CBR uses an incremental approach to sustained learning, since a new experience is retained each time a problem has been solved, making it available for future problems.

In the next post, we will see how to solve problems using CBR.