Abstract:
|
Statistical analysis for the purpose of prediction is preferably accompanied by uncertainty quantification, often in the form of prediction intervals. In many applications, deep learning approaches have been extensively shown to provide accurate point predictions. However, the generation of prediction intervals within conventional deep learning models calls for diligent adaptation, development, testing, and implementation. To this end, we propose a novel deep learning model which provides both accurate point predictions and prediction intervals, through empirical score minimization of proper scoring rules for interval forecasts. In simulation studies and real data applications, we demonstrate the efficacy of the novel deep learning model against traditionally used methods.
|