Using only retained subsets of rules, learning process was performed again with the input values of the stored learning data, and the consequent values were rearranged. Using the new template set increased accuracy of identification of the two states.