A Framework for Evaluation and Identification of Predictive Dynamic Systems Models

 

A Framework for Evaluationand Identification of Predictive Dynamic Systems ModelsThe goal of this project is to develop novel dynamic systems models that can be used for robust short term predictions of physiological signals. Most traditional modeling and prediction methods are based on minimizing the prediction error. In many physiological settings the absolute or mean square prediction error is not as important as the qualitative characteristics of the prediction (e.g. breaching a clinical threshold for a minimum duration). Recognizing this, our models are developed to optimize the ability to detect clinically relevant events. multivariate dynamical models are considered to address the complex directionality between physiological variables. The long term potential of this project is to develop novel control systems that can provide prophylactic intervention to prevent critical patient outcomes.
Researcher: Jungyoon (JY) Kim