This paper presents the details of the design of the algorithm 2G, which is based on techniques of pattern recognition and decision trees, and describes original contributions that consist in: the use a method different from the existing ones in the stage of values discretización, eliminating redundancies and taking into account values for which there are overlapping classes, considering them, at this stage, as values grouped into 'virtual classes' that allows them to have representation on the set of values involved in the selection of attributes; the application of the criterion for minimizing the amount of attributes to construct the corresponding decision tree, which results in the decreasing number of conditions required to evaluate the instances for which there is not known the class to which they belong; and to allow the classification of new examples with no explicit rules derived from the rules generated, with the only restriction that these rules must belong to the same class. In addition, it also includes classification results obtained with the 2G algorithm using cross-validation, which almost always improves the results reported in the literature.
|Original language||Spanish (Mexico)|
|Title of host publication||World Scientific Proceedings Series on Computer Engineering and Information ScienceComputational Intelligence in Business and Economics|
|Subtitle of host publication||2G: A classification algorithm based on pattern recognition and decision trees|
|Place of Publication||Barcelona, España|
|Number of pages||6|
|State||Published - 1 Jul 2010|