Bad and good errors: value-weighted skill scores in deep ensemble learning

TL;DR

This paper introduces a novel forecast verification method emphasizing the importance of prediction value, and applies it to deep ensemble learning for binary classification in pollution, space weather, and stock forecasting.

Contribution

It proposes a new value-weighted skill score and a deep ensemble learning approach optimized by these scores for improved forecast assessment.

Findings

Enhanced forecast accuracy in pollution, space weather, and stock prediction.

The value-weighted skill scores better reflect the real impact of forecast errors.

Deep ensemble models outperform traditional methods in the tested applications.

Abstract

In this paper we propose a novel approach to realize forecast verification. Specifically, we introduce a strategy for assessing the severity of forecast errors based on the evidence that, on the one hand, a false alarm just anticipating an occurring event is better than one in the middle of consecutive non-occurring events, and that, on the other hand, a miss of an isolated event has a worse impact than a miss of a single event, which is part of several consecutive occurrences. Relying on this idea, we introduce a novel definition of confusion matrix and skill scores giving greater importance to the value of the prediction rather than to its quality. Then, we introduce a deep ensemble learning procedure for binary classification, in which the probabilistic outcomes of a neural network are clustered via optimization of these value-weighted skill scores. We finally show the performances of this approach in the case of three applications concerned with pollution, space weather and stock prize forecasting.