Decomposition

First we introduce the terms model and parameter . A model is a description of the state of some data given in a language or as a graphical representation, for instance as the structure of a Bayesian network. The parameters for a model are data for the model that can be learned from a database, for instance for a Bayesian network, the parameters are the probabilities on the edges. Then we can decompose the Data Mining algorithms into:

• Model representation - the model should be represented in a language.
• Model evaluation - the model can be evaluated according to metrics as novelty, utility, validity, accuracy or understandability.
• Search method - There are two ways of searching the data - parameter search and model search . The first is to find parameters from the data for a fixed model, the second to evaluate different possible models according to the database.

We distinguish between supervised data mining, where the user has to identify a specific goal or weigh features before the operation can begin, and unsupervised data mining, which does not require any interaction with the user.

Below, a method for Data Mining which will be used in illustrating integration of Data Mining and Case-Based Reasoning is outlined.

  

Figure 3.2: An Example Bayesian Network.