Estimating neuronal variable importance with Random Forest

We describe a novel application of a new method for data mining, Random Forest, for the analysis of data sets acquired with neuronal ensemble recording methods. Random Forest is used here to measure the relative importance of each input variable in the data. The technique is fast and can greatly reduce the number of variables with little compromise, especially for highly redundant data like neural ensemble spike trains. It also naturally preserves the identifiability of the original Information source unlike other techniques, for example the principal component analysis that mixes up the content of the variables irrecoverably and projects it into different set of variables.