When a multilayer perceptron (MLP) is trained with the split-complex backpropagation (SCBP) algorithm, one observes a relatively strong dependence of the performance on the initial values. For the effective adjustments of the weights and biases in SCBP, w