Uncertainty Modelling in Deep Networks: Forecasting Short and Noisy Series

TL;DR

This paper explores uncertainty modeling in deep networks for financial time series forecasting, demonstrating improved accuracy and a mechanism to discard low-confidence predictions to enhance user experience.

Contribution

It introduces a heteroscedastic deep learning approach for uncertainty estimation and a method to filter out low-confidence forecasts in financial applications.

Findings

Higher accuracy than baseline models

Effective filtering of low-confidence predictions

Improved user experience through uncertainty-based filtering

Abstract

Deep Learning is a consolidated, state-of-the-art Machine Learning tool to fit a function when provided with large data sets of examples. However, in regression tasks, the straightforward application of Deep Learning models provides a point estimate of the target. In addition, the model does not take into account the uncertainty of a prediction. This represents a great limitation for tasks where communicating an erroneous prediction carries a risk. In this paper we tackle a real-world problem of forecasting impending financial expenses and incomings of customers, while displaying predictable monetary amounts on a mobile app. In this context, we investigate if we would obtain an advantage by applying Deep Learning models with a Heteroscedastic model of the variance of a network's output. Experimentally, we achieve a higher accuracy than non-trivial baselines. More importantly, we introduce a mechanism to discard low-confidence predictions, which means that they will not be visible to users. This should help enhance the user experience of our product.