A0455
Title: Robust forecasting with machine learning
Authors: Christophe Croux - KU Leuven (Belgium) [presenting]
Abstract: The impact of outliers on time series prediction is discussed. The use of ARMA models for robust estimation and prediction is well studied. For nonlinear models, however, machine learning methods are a suitable alternative. Popular machine learning methods such as random forests, XGBoost, and LightGBM can also be used in a time series context. Their predictive performance is examined in the presence of outliers. Moreover, it is investigated how these methods can be made more robust by changing the loss function and adding a filtering step.
