Confidence measure estimation in dynamical systems model input set selection

An information-theoretic input selection method for dynamical system modeling is presented that qualifies the rejection of irrelevant inputs from a candidate input set with an estimate of a measure of confidence given only finite data. To this end, we introduce a method of determining the spatial interval of dependency in the context of the modeling problem for bootstrap mutual information estimates on dependent time-series. Additionally, details are presented for determining an optimal binning interval for histogram-based mutual information estimates.