A database may be consistent or inconsistent. The data is inconsistent if some objects with all the same condition attributes have a different classification, otherwise it is consistent. Thus, in the inconsistent case, the data mining system is not capable of separating some objects from each other when regarding the the condition attributes. Inconsistencies may arise due to lacking information in the database, noise or measurement errors.
The knowledge may be either deterministic or
non-deterministic. Deterministic knowledge is knowledge which is
believed to be certain, and could for instance be represented by
definite rules, see Section 
. Definite rules say
that something is right for all objects, whereas non-deterministic
rules may say that something is usually so. In most cases
deterministic rules will be created for consistent data, and
non-deterministic rules for inconsistent data. Because of this, it is
quite common to mix the concepts deterministic and consistent freely.
The representations we present here may be deterministic or indeterministic in the way that they may not be correct for all data in the database, but cover only the most common cases. First we present the probably most usual way of representing knowledge, propositional representations. Then we discuss the abilities of first order logic, before mentioning a fundamentally different knowledge representation form called neural net.