Constructing Simpler Decision Trees from Ensemble Models Using Fourier Analysis.
Byung Hoon Park, Hillol Kargupta · 2002
Ensemble learning is frequently used for classification and other related applications in data mining. It generates multiple models and produces the final classification by aggregating the outputs of the different models in the ensemble. However, large ensembles are often hard to interpret and difficult to translate into action-able knowledge. This paper considers the construction of a decision tree from the Fourier spectrum of an ensemble model within a user-defined range of errors. The Fourier spectrum of an ensemble of decision trees retains all the necessary information that can be used to construct a simpler "informative" decision tree. This approach can be effectively used for building ensemble-based classifiers from both static data sets and data streams.