S4: Neural Networks for Time Series Forecasting

Grzegorz Dudek

Czestochowa University of Technology, Poland

Abstract

Time series forecasting is an important problem in many domains such as business, industry, economics, engineering, and science. It is relevant and challenging as the time series expressing real-world phenomena are in many cases very complex and stochastic, including trends, multiple seasonality, and significant random fluctuations. Therefore, time series forecasting is an active research area that has received a considerable amount of attention from researchers and practitioners for many years. 

Over the past few decades, neural networks have been successfully used in this field due to their ability to capture different patterns and high expressive power to solve non-linear stochastic forecasting problems. However, in practice, it is quite challenging to properly determine an appropriate architecture and parameters of neural networks as well as the training process and time series representation so that the resulting forecasting model can achieve sound performance for both learning and generalization. Practical applications of neural forecasting models bring additional challenges, such as dealing with big, missing, distorted, and uncertain data. In addition, interpretability, explainability, and causality are paramount qualities that neural methods should aim to achieve if they are to be applied in practice.

This Special Session focuses on neural models and their application in a diverse range of forecasting areas and problems including probabilistic and multi-step forecasting. The papers are expected to report substantive results on a wide range of neural forecasting models (variants of MLP, CNN, RNN), discussing their architectures and training procedures, the conceptualization of a problem, time series representation, feature engineering, critical comparisons with existing techniques, and interpretation of results. Specific attention will be given to recently developed neural forecasting models including deep and hybrid solutions.