Contextual classifier ensembles

Individual classifiers do not always yield satisfactory results. In the field of data mining, failures are mainly thought to be caused by the limitations inherent in the data itself, which stem from different reasons for creating data files and their various applications. One of the proposed ways of dealing with these kinds of shortcomings is to employ classifier ensembles. Their application involves creating a set of models for the same data file or for different subsets of a specified data file. Although in many cases this approach results in a visible increase of classification accuracy, it considerably complicates, or, in some cases, effectively hinders interpretation of the obtained results. The reasons for this are the methods of defining learning tasks which rely on randomizing. The purpose of this paper is to present an idea for using data contexts to define learning tasks for classifier ensembles. The achieved results are promising.