|
Abstract:
|
Tests may be sequenced as a method of improving classification accuracy while minimizing operational cost. Here, operational cost refers to the expense of operating a combination of tests, such as financial cost or the time required to complete the sequence of tests. Currently, there exists little guidance to determine the number of tests to be sequenced in order to achieve the desired testing accuracy with an associated operational cost. This work presents a weighted function of sequence accuracy and operational cost which may be used to determine the number of tests to sequence for a diagnostic task. Computational formulas are presented for sequencing accuracy, as represented by true and false positives, as well as operational cost. Weighting is provided for trade-offs between accuracy and operational cost, as well as misclassification costs and class prevalence. A method is presented to compute the variance of this function so that inferences such as a confidence interval or hypothesis test could be made. Simulated results demonstrate the effects of individual test accuracy, prevalence, correlation, and operational and misclassification costs on the optimal sequence.
|
