User Traffic Time Series Uncertainty Estimation with Deep Laplace Ensemble

User traffic time series prediction is a common problem in various domains. However, in many cases, we want to know not only the predicted value of user traffic, but also the uncertainty of that predicted value, so that we can make appropriate decisions. Inspired by the deep ensemble and laplace approximation, we propose a new approach for user traffic uncertainty prediction, which combines deep ensemble and laplace approximation and leverages the flexibility of deep learning architectures in capturing diverse data representations while incorporating the concept of deep ensemble and laplace approximation to model the inherent uncertainty in the predictions. We conduct experiments using actual user traffic datasets, and the results show that our model has similar performance in the accuracy of prediction results as the vanilla model, but outperforms in robustness. Beyond that, we design a new method to show that our model can really give the uncertainty in OOD dataset.