Two-Stage Neural Network Classification Structure
The two-stage neural network classification unit including S1 tan-sigmoid transfer functions, denoted by F1, in the first neuron layer and one linear transfer function, denoted by F2, in the second neuron layer is shown in Figure 2. R and S1 represented the total number of attributes (including the combined attributes) and the number of hidden neurons in the first neuron layer, respectively. W1, B1 and W2, B2 represented hidden neurons of the weights and biases in the first and second neuron layers, respectively. In our research, R signified the 7 attributes (where five were independent attributes and two were combined attributes), and S1 was set to include 40 hidden neurons. W1 and W2 were the weights of the (40×7) matrix and (1×40) matrix, respectively; and B1 and B2 were the biases of the (40×1) matrix and (1×1) matrix (which is a scalar), respectively.