In order to illustrate this algorithm, suppose we have a large database of one million ED records over a two-year span. This database contains roughly 1370 records a day. Suppose we treat all records within the last 24 hours as "recent" events. In addition, we can build a baseline data set out of all cases from exactly 35, 42, 49, and 56 days prior to the current day. We then combine the recent and baseline data to form a record subset called DB, which will have approximately 5000 records. The algorithm proceeds as follows. We first consider all possible one-component rules. For every possible feature—value combination, obtain the counts Crecent and Cbaseiine from the data set DB. As an example, suppose the
Was this article helpful?