General noise SVRs and uncertainty intervals

Abstract

Building uncertainty estimates is still an open problem for most machine learning regression models. On the other hand, general noise–dependent cost functions have been recently proposed for Support Vector Regression, SVR, which should be more effective when applied to regression problems whose underlying noise distribution follows the one assumed for the cost function. Taking this into account, we first propose a framework that combines general noise SVR models trained by Naive Online R Minimization Algorithm, NORMA, optimization with uncertainty interval estimates for their predictions. We then provide the theoretical details required to implement this framework for several noise distributions and carry out experiments, whose results show an improvement over the ones obtained by classical ɛ-SVR and also support the hypothesis that the model and error intervals with the noise distribution assumption closest to the real one yield the best results. Finally, and in accordance with the principle of reproducible research, we make the implementations developed and the datasets employed in the experiments publicly and easily available.