Set-Covering Classification

Set-covering classification is suitable for classification problems in which the solutions (causes) evoke particular symptoms (effects) — possibly via intermediate states — with a relatively high reliability. In the simplest form the knowledge representation consists of observations, solutions and rules of the form: solution causes observation (S → O). The probability of a given solution or group of solutions is greater, the more observations are explained, i.e., covered, according to its rules and the smaller the number of non-observed features which can be derived from it. The basic structure is illustrated in When set-covering classification is used for fault-finding it is called classification with fault models. Two examples of fault models from the technical and medical fields, respectively, are shown in Figs. 17.2 and 17.3. Set-covering is not only suitable for fault-finding, however, but for all domains in which the solutions may be described by characteristic sets of problem features.