Dynamic and static approaches to clinical data mining

摘要

In sequential diagnosis, the usefulness of a test can be assessed only in the context of a chosen diagnostic strategy, and depends on the evidence provided by previous test results. Choosing the most useful test at each stage of the evidence-gathering process therefore requires a dynamic approach to data analysis. An implementation of such an approach in an intelligent program for sequential diagnosis based on the evidence-gathering strategies used by doctors is described. On the other hand, a static approach to data analysis is appropriate in the discovery of knowledge required, for example, to explain or justify a diagnosis by identifying the most important findings, both positive and negative, on which the diagnosis is based. An algorithm for the discovery of features which always provide evidence in favour of, or against, a diagnosis selected by the data miner is presented. Dominance relationships among features in the data set are also discovered such that if one feature dominates another, it always provides more evidence in favour of the diagnosis, or less evidence against it.