Decision support using machine learning: Towards intensive care unit patient state characterization

We present a framework for the study of real-world time-series data using supervised Machine Learning techniques. This methodology has been developed to suit the needs of data monitoring in Intensive Care Unit, where data are highly heterogeneous. It is based on the windowed processing and monitoring of model characteristics, in order to detect changes in the model. These changes are considered to reflect the underlying systems' state transitions. We apply this framework after specializing it, based on field knowledge and ex-post corroborated assumptions.