Neural networks are still frustrating tools in the data mining arsenal. They exhibit excellent modelling performance, but do not give a clue about the structure of their models. We propose a methodology to explain the classification obtained by a multilayer perceptron.
View Article and Find Full Text PDF