The biological data collected from intensive care units contain signal and noise. To extract information that will be useful for predicting or discriminating the cases likely to develop an acute hypotensive episode (AHE), we begin by applying a spline-based smoothing method to the observed mean arterial pressure (MAP) curves. The coefficients of the fitted spline model form a discretization matrix of the continuous MAP curves. A rank-based discriminant analysis and a cross-validation method are developed to find the best prediction subset in the training set. The selected best subsets are used to predict AHE in the test sets. This work is from participation of PhysioNet/Computers in Cardiology Challenge 2009: Predicting Acute Hypotensive Episodes.
展开▼
