Learning Uncertainty with Artificial Neural Networks for Improved Remaining Time Prediction of Business Processes

Artificial neural networks will always make a prediction, even when\ncompletely uncertain and regardless of the consequences. This obliviousness of\nuncertainty is a major obstacle towards their adoption in practice. Techniques\nexist, however, to estimate the two major types of uncertainty: model\nuncertainty and observation noise in the data. Bayesian neural networks are\ntheoretically well-founded models that can learn the model uncertainty of their\npredictions. Minor modifications to these models and their loss functions allow\nlearning the observation noise for individual samples as well. This paper is\nthe first to apply these techniques to predictive process monitoring. We found\nthat they contribute towards more accurate predictions and work quickly.\nHowever, their main benefit resides with the uncertainty estimates themselves\nthat allow the separation of higher-quality from lower-quality predictions and\nthe building of confidence intervals. This leads to many interesting\napplications, enables an earlier adoption of prediction systems with smaller\ndatasets and fosters a better cooperation with humans.\n