Classification 1: House and Purple Finches

The first network that was built was trained to distinguish between the House and Purple finches. These birds were chosen due to the similarity of their calls, and the variation of the vocalizations for each species observed within the data sets. The next figure shows a comparison of the frequency spectrum of the vocalizations of these two species.

The network that was built consisted of a single hidden layer, between and input and output layers. The input layer supplied the twenty-five vector elements as inputs. The hidden and output layers both contained four and two neurons respectively, and used hyperbolic tangent as their discriminating function. This network formed the basis for all further testing. This simple network trained very quickly and provided one hundred percent accurate categorization.