Importance of the Transformer Models in Forecasting and Classification of Time Series

Abstract

Transformer models have brought significant advancements in time series analysis due to their ability to capture long-term dependencies. This study examines the fundamental structure, strengths, and limitations of transformer models (deep learning architectures) in time series modeling. It presents various transformer variants for time series and discusses their applications in forecasting, classification, and anomaly detection, as well as the explainability and interpretability.