Enhancing Automated Test Selection in Probabilistic Networks

In diagnostic decision-support systems, test selection amounts to selecting, in a sequential manner, a test that is expected to yield the largest decrease in the uncertainty about a patient’s diagnosis. For capturing this uncertainty, often an information measure is used. In this paper, we study the Shannon entropy, the Gini index, and the misclassification error for this purpose. We argue that the Gini index can be regarded as an approximation of the Shannon entropy and that the misclassification error can be looked upon as an approximation of the Gini index. We further argue that the differences between the first derivatives of the three functions can explain different test sequences in practice. Experimental results from using the measures with a real-life probabilistic network in oncology support our observations