Efficient Case Based Feature Construction for Heterogeneous Learning Tasks
Ingo Mierswa, Michael Wurst · 2006
1 Introduction Many inductive learning problems cannot be solved accurately by using theoriginal feature space. This is due to the fact that standard learning algorithms cannot represent complex relationships as induced for example by trigonometricfunctions. For example, if only base features X1 and X2 are given but the targetfunction depends highly on Xc = sin(X1 *X2), the construction of the feature Xcwould ease learning- or is necessary to enable any reasonable predictions at all