An empirical measure of element contribution in neural networks

Abstract

A frequent complaint about neural net models is that they fail to explain their results in any useful way. The problem is not a lack of information, but an abundance of information that is difficult to interpret. When trained, neural nets will provide a predicted output for a posited 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 relative 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.Index Terms-Clustering methods, hidden element contribution, input element contribution, measurement index, neural network architecture.

Simplifying artificial neural network models of river basin behaviour by an automated procedure for input variable selection

“…The influence of variables is measured by the relation between the weights associated with each input node and the sum of all synaptic weights. Studies by Mak and Blanning (1998) and Guha et al (2005) also give variants of this method. The third algorithm is based on the calculation of partial derivatives of ANN outputs with respect to the inputs (see Dimopoulos et al, 1995).…”

Rule extraction from expert heuristics: A comparative study of rough sets with neural networks and ID3

“…To address the rates of change across the hidden layer, Mak and Blanning [22] developed the following index to assess the impact of input i on output k:…”

“…In total 233 cases were collected, 105 cases were used for training and 128 cases for testing. Details of the knowledge acquisition process are found in [22,23].…”

“…A neural network of one hidden layer with three hidden units was constructed to analyze the data. This structure was adopted because it was the most parsimonious neural network architecture used for modeling the data, and addition of additional elements in the hidden layer did not result in increase in accuracy [22]. Table 7 shows the robustness and predictive ability of rough sets.…”

“…Examples include the work of Garson [11], Yoon et al [66], and Mak and Blanning [22]. These indices are measures showing the relative importance of the inputs in contributing to the outputs.…”