2.3. Analysis of trained neural network
It is possible to carry out an analysis of a trained neural network, in order to determine the level of influence of each input variable on output nodes. Based on the work of Olden et al. (2004) which builds on that of Olden and Jackson (2002), this was carried out by determining the product of the connections weights between input/hidden layer 1, hidden layer 1/hidden layer 2 and hidden layer 2/output nodes for each combination of input neuron and output neuron, with each combination of input, hidden and output nodes summed across combinations of input and output nodes. The absolutes of the resultant values are normalised within the range [0, 1]. Due to the fact that leave-one-out analysis was used and that a total of 127 networks were trained, the means of input parameter influence over all networks were calculated. The values given from this analysis tell us which input parameters are more influential than others
in terms of predicting individual outputs, although they do not give a direct numerical relationship between inputs and outputs.