Interpretation and Analysis of Input Selection Approaches in Distance Space

In this paper five different model free input selection approaches are summarized. While all the criteria are based on geometrical properties of the data, it is possible to visualize most of the approaches in the here called ´Distance Space´. The five discussed approaches for input selection are interpreted and analyzed in the Distance Space scatter plot. The influence of noise and wrong chosen inputs to the Distance Space is outlined. With the chosen search strategy, the input space is changed. Afterward the best input combination is selected. Several dynamic systems are used to generate data sets, in order to compare the different approaches.