Extracting Temporal Rules from Medical Data

The work presented in this paper is the application of temporal data mining for discovering hidden knowledge from medical dataset. Medical data is temporal in nature and therefore conventional data mining techniques are not suitable. This dataset contains medical records of pregnant mothers. The structure of these medical records is chain of observations taken at different times. In each observation, a set of clinical parameter is saved by midwives. The aim of this paper is mining temporal relational rules from this set of temporal interval data that can be used in early prediction and of risk in the patients. In the first part of this study a pre-processing technique is used to produce temporal interval data from primary structure of medical records. Three different analyses are studied in preprocessing phase due to the complexity of medical records and differences in the sequence of observed symptoms in various diseases. In the next phase the mining algorithm is used to extract temporal rules. The base of this algorithm is Allen's temporal relationship theory. The rules are represents as directed acyclic graphs. The generated rules can be used in diagnosis of risk full phenomena in antenatal care. Mining medical data for this case becomes very significant as many of the current maternal deaths or birth of premature newborns might be prevented by prediction and early detection of high risk patients.