[논문 리뷰] Time Series Analysis and Modeling to Forecast: a Survey

Fatoumata Dama, Christine Sinoquet|arXiv (Cornell University)|2021. 03. 31.

Stock Market Forecasting Methods참고 문헌 138인용 수 30

한 줄 요약

A comprehensive survey of time series forecasting covering decomposition, stationary tests, linear/nonlinear models, and deep learning, with practical implementation guidance.

ABSTRACT

Time series modeling for predictive purpose has been an active research area of machine learning for many years. However, no sufficiently comprehensive and meanwhile substantive survey was offered so far. This survey strives to meet this need. A unified presentation has been adopted for entire parts of this compilation. A red thread guides the reader from time series preprocessing to forecasting. Time series decomposition is a major preprocessing task, to separate nonstationary effects (the deterministic components) from the remaining stochastic constituent, assumed to be stationary. The deterministic components are predictable and contribute to the prediction through estimations or extrapolation. Fitting the most appropriate model to the remaining stochastic component aims at capturing the relationship between past and future values, to allow prediction. We cover a sufficiently broad spectrum of models while nonetheless offering substantial methodological developments. We describe three major linear parametric models, together with two nonlinear extensions, and present five categories of nonlinear parametric models. Beyond conventional statistical models, we highlight six categories of deep neural networks appropriate for time series forecasting in nonlinear framework. Finally, we enlighten new avenues of research for time series modeling and forecasting. We also report software made publicly available for the models presented.

연구 동기 및 목표

• Provide an informed, self-contained overview of time series forecasting from preprocessing to prediction.
• Unify decomposition, stationarity analysis, and modeling under a broad methodological spectrum.
• Bridge classical statistical methods and modern deep learning approaches for practitioners.
• Highlight available software implementations and identify avenues for future research.

제안 방법

• Present a unified framework from time series preprocessing to forecasting.
• Review stationary concepts and tests, including strong/weak stationarity.
• Describe time series decomposition into deterministic and stochastic components.
• Detail three major linear models, two nonlinear extensions, and five nonlinear model categories.
• Discuss deep learning approaches for time series forecasting in a nonlinear framework.
• Offer guidance on model evaluation and provide software implementations.

실험 결과

연구 질문

• RQ1 What are the main preprocessing, decomposition, and forecasting steps in time series analysis?
• RQ2 What linear and nonlinear models are suitable for forecasting, and how do stationarity and nonstationarity affect them?
• RQ3 How do deep neural networks fit into time series forecasting alongside traditional methods?
• RQ4 What software resources are available to implement the surveyed models and tests?
• RQ5 What are the open research directions and future prospects in time series forecasting?

주요 결과

• This is, to the authors’ knowledge, the first comprehensive survey dedicated to forecasting in time series.
• The paper provides a unified presentation of decomposition frameworks and of linear/nonlinear models.
• It explains the links between stationarity and linearity to help nonspecialists, bridging theory and practice.
• The survey spans traditional approaches and modern deep learning adaptations for forecasting.
• It includes guidance on implementation through publicly available R and Python resources and outlines future research directions.