AN EMPIRICAL MEASURE OF ELEMENT CONTRIBUTION IN NEURAL NETWORKS

A frequent complaint about neural net models is that they fail to expl ain their results in any useful way, The problem is not a lack of info rmation, but an abundance of information that is difficult to interpre t. When trained, neural nets will provide a predicted output for a pos ited input, and they can provide additional information in the form of interelement connection strengths. But this latter information is of little use to analysts and managers who wish to interpret the results they have been given. In this paper, we develop a measure of the relat ive importance of the various input elements and hidden layer elements , and we use this to interpret the contribution of these components to the outputs of the neural net.