Temporal knowledge discovery in time-series medical databases based on fuzzy-rough reasoning

Since hospital information systems have been introduced in large hospitals, a large amount of data, including laboratory examinations, have been stored as temporal databases. The characteristics of these temporal databases are outlined. (1) Each record is inhomogeneous with respect to time-series, including short-term effects and long-term effects. (2) Each record has more than 1000 attributes when a patient is followed for more than one year. (3) When a patient is admitted for a long time, a large amount of data is stored in a very short term. Even medical experts cannot deal with these large databases, and the interest in mining some useful information from the data are growing. The article introduces a combination of extended moving average method and rule induction method, called CEARI to discover new knowledge in temporal databases. An extended moving average method is used for preprocessing, to deal with the irregularity of temporal data. Using several parameters for time-scaling, given by users, this moving average method generates a new database for each time scale with summarized attributes. Then, the rule induction method is applied to each new database with summarized attributes. This CEARI is applied to three medical datasets, the results of which show that interesting knowledge is discovered from each database