Computation of a Sucient Condition for System Input Redundancy
Stelios E. Papadakis, Vassilis G. Kaburlasos · 2008
The calculation of an optimal subset of inputs from a set of candidate ones is known in the bibliography of system modeling as the input (or feature) selection problem. In this work we introduce a remarkable attribute of the FLR classifier: it's capacity to identify re- dundant system inputs, from a set of input/output data. The proposed approach is applicable beyond R N on any lattice ordered data set L N , which may include disparate types of data. Also, the proposed approach can deal with populations of data instead of crisp data vectors. Finally, it is highlighted that proposed approach can be employed for designing models with simple structure and significant performance. The method is successfully applied here on two well known real world classification problems, identifying redundant inputs and inducing FLR classifiers with simple structure and favorable classification performance.