P. Dagum, A. Galper, and A. Seiver
We explore the use of the delay-coordinateembedding technique to automatically construct and use probability forecast models. The technique offers distinct advantages, including coherent handling of noise and continuous data, over previously explored  probability forecast methods employing Bayesian belief networks. We discuss the relationship between the dynamical systems and the probabilistic reasoning approaches to probability forecasting. Finally, we apply the dynamical systems method to a multivariate time series of physiologic measurements of an infant with adult respiratory distress syndrome (ARDS).