Automated Model Selection
Many algorithms have parameters that should be set by the user.
For most machine learning algorithms parameter setting is a
non-trivial task that influence knowledge model returned
by the algorithm.
Parameter values are usually set approximately
according to some characteristics of the target problem,
obtained in different ways.
The usual way is to use background knowledge about the target
problem (if any) and perform some testing experiments.
The paper presents an approach to automated model selection
based on local optimization that uses an empirical evaluation
of the constructed concept description to guide the search.
The approach was tested by using the inductive concept learning
system Magnus Assistant on several problem domains.