A Comprehensive Survey of Deep Learning for Time Series Forecasting: Architectural Diversity and Open Challenges

TL;DR

This survey maps the architectural renaissance in time series forecasting, tracing the shift from traditional statistics and early DL models to a spectrum of backbones—MLPs, CNNs, RNNs, GNNs, Transformers, diffusion models, and Mamba SSMs—and toward foundation models. It highlights how recent findings, such as linear models outperforming Transformers in long-term forecasting, have sparked renewed interest in non-transformer architectures and hybrid approaches, while still recognizing the continued relevance of Transformers in many settings. The authors organize current advances around seven focal challenges—channel dependency, distribution shift, causality, feature extraction, model combination, interpretability, and spatio-temporal forecasting—and present practical handling methods and trend analyses. They emphasize the need for robust, interpretable, and generalizable TSF systems and offer a structured roadmap for researchers entering or advancing in TSF, including foundation-model-oriented directions and diffusion-based uncertainty modeling. The paper concludes that a balanced ecosystem of backbone models, domain-aware feature engineering, and principled evaluation will drive practical, scalable, and trustworthy TSF solutions across domains.

Abstract

Time series forecasting is a critical task that provides key information for decision-making. After traditional statistical and machine learning approaches, various fundamental deep learning architectures such as MLPs, CNNs, RNNs, and GNNs have been developed. However, the structural limitations caused by the inductive biases of each deep learning architecture constrained their performance. Transformer models, which excel at handling long-term dependencies, have become significant architectural components for time series forecasting. However, recent research has shown that alternatives such as simple linear layers can outperform Transformers. These findings have opened up new possibilities for using diverse architectures, ranging from fundamental deep learning models to emerging architectures and hybrid approaches. In this context, architectural modeling of time series forecasting has now entered a renaissance. This survey not only provides a historical context for time series forecasting but also offers comprehensive and timely analysis of the movement toward architectural diversification. By comparing and re-examining deep learning models, we uncover new perspectives and present recent trends, including hybrid, diffusion, Mamba, and foundation models. By focusing on the inherent characteristics of time series data, we also address open challenges that have gained attention in time series forecasting, such as channel dependency, distribution shift, causality, and feature extraction. These contributions help lower entry barriers for newcomers by providing a systematic understanding of the diverse research areas in time series forecasting (TSF), while offering seasoned researchers broader perspectives and new opportunities through in-depth exploration of TSF challenges. (Shortened due to arXiv's 1,920-character limit. Full version in the paper.)

Figures (18)

• Figure 1: Overview of This Survey
• Figure 2: Number of Top-tier AI and ML Conference Papers on Time Series Forecasting
• Figure 3: Evolution of Time Series Forecasting Models
• Figure 4: Overview of Latest Open Challenges in Time Series Forecasting
• Figure 5: Properties of Time Series Data