EconBase
← All papers

Prediction intervals for Deep Neural Networks

Tullio Mancini, Hector Calvo-Pardo, Jose Olmo

arXiv 8 Oct 2020 · Statistics — Machine Learning · 3 citations (OpenAlex)

arXiv:2010.04044 · PDF · DOI · OpenAlex · Extracted main text

Abstract

The aim of this paper is to propose a suitable method for constructing prediction intervals for the output of neural network models. To do this, we adapt the extremely randomized trees method originally developed for random forests to construct ensembles of neural networks. The extra-randomness introduced in the ensemble reduces the variance of the predictions and yields gains in out-of-sample accuracy. An extensive Monte Carlo simulation exercise shows the good performance of this novel method for constructing prediction intervals in terms of coverage probability and mean square prediction error. This approach is superior to state-of-the-art methods extant in the literature such as the widely used MC dropout and bootstrap procedures. The out-of-sample accuracy of the novel algorithm is further evaluated using experimental settings already adopted in the literature.

Citation extraction

65
references
0
in-text mentions
0
distinct cited
1
self-citations
16,678
main-text words

appendix boundary found by appendix_titled_section at “Appendix: Random weight initialization” · 98% of the source is main text. Read the extracted text to check this.