Issue No. 02 - March/April (1998 vol. 13)
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/5254.671090
Function decomposition decomposes a dataset to an equivalent hierarchy of less complex datasets. In this article, we show that function decomposition can discover a hierarchy of new features and construct features to be added to the original dataset. An advantage of function decomposition is its ability to identify appropriate subsets of existing features and discover functions that map each subset to a new feature. These functions are induced from examples and are not predefined in any way.
feature transformation, feature construction, feature discovery, function decomposition, hierarchy of features, classification
Blaz Zupan, Janez Demsar, Marko Bohanec, Ivan Bratko, "Feature Transformation by Function Decomposition", IEEE Intelligent Systems, vol. 13, no. , pp. 38-43, March/April 1998, doi:10.1109/5254.671090