The receiver operating characteristic function as a tool for uncertainty management in artificial neural network decision-making

A technique for enhancing artificial neural network (ANN) performance is presented. This technique uses receiver operating characteristic methodology to adjust the operating threshold values of ANN output classification processing units to account for both prevalence differences between training cases and real-world cases, and for unequal costs incurred with false positive and false negative classifications. The basic task is to incorporate knowledge of prevalence and error costs when making individual decisions using trained neural networks. The technique is illustrated with a back-error propagation neural network.<<ETX>>