A full explanation facility for a MLP network that classifies low-back-pain patients

This paper presents a full explanation facility that has been developed for a standard MLP network, with binary input neurons that performs a classification task. It is shown how an explanation for any input case is represented by a non-linear ranked data relationship of key inputs, in both text and graphical forms. Using the facility, the knowledge that the MLP has learned can be represented by average ranked class profiles or as a set of rules induced from all training cases. The full explanation facility discovers the MLP knowledge bounds by finding the hidden layer decision regions containing correctly classified training examples. Novel inputs are detected by the explanation facility, on an input case-by-case basis, when the case is positioned in a decision region outside the knowledge bounds. Results using the facility are presented for a real-world MLP network that classifies low-back-pain patients.