Analysis of ICU Patients Using the Time Series Knowledge Mining Method

Time oriented data presents a more detailed description of problems, while presenting challenges in the computational needs for a successful analysis, in which the time is explicitly analyzed. Commonly temporal datasets are converted into a static representation and being analyzed by common static data mining methods, such as decision trees. Abstracting time series into time intervals, using temporal abstraction, enables to analyze the data explicitly along time. We apply here a mining method, which discovers partially ordered coinciding time intervals, consisting on the Time Series Knowledge Mining (TSKM) method, presented by Mörchen [2006] on temporal data of intensive care patients, using human defined and two types of data driven temporal descretization methods as a preprocessing step. The Persist discretization method results with the best knowledge discovery outcome.