On the Relation between Data Mining and Knowledge Mining Comments on the Book by Professor J.-A. Mueller and F. Lemke 'Self-Organizing Data Mining'

摘要

The algorithms for solving interpolation problems of artificial intelligence that do not involve constructing mathematical models can conventionally be referred to data mining. Such algorithms exhibit a high accuracy on the samples used for their construction. To knowledge mining, all algorithms using construction of mathematical models should be referred. By applying inductive search modeling methods, in particular, by choosing a suitable criterion, both a nonphysical model, which exhibits the highest accuracy on the test data sample, and a physical model, which has the generalization property (i.e., ensures a high accuracy on other data samples) can be obtained. Generalization is a characteristic property of the clustering and of the models obtained with the use of the equilibrium or unbiasedness criterion. A repeated application of modeling makes it possible to perform a Kalman-type filtering of a noise that cannot be formalized and measured.