Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TimeRecipe: A Time-Series Forecasting Recipe via Benchmarking Module Level Effectiveness

Zhiyuan Zhao, Juntong Ni, Shangqing Xu, Haoxin Liu, Wei Jin, B. Aditya Prakash

TL;DR

TimeRecipe addresses the disconnect between holistic benchmarking and the need for understanding module-level effectiveness in time-series forecasting. It proposes a canonical, module-centered framework to evaluate pre-processing, embedding, and feed-forward components, conducting over 10,000 experiments across univariate/multivariate settings, short/long horizons, and diverse datasets. Key contributions include (i) a comprehensive modular benchmark that uncovers architectures outperforming state-of-the-art methods, (ii) correlations linking module performance to data properties, and (iii) a training-free LightGBM-based toolkit for practical architecture selection. The work demonstrates that exhaustive modular exploration yields superior performance and that design choices should align with intrinsic data characteristics, offering actionable guidance and a path toward adaptive AutoML for time-series forecasting.

Abstract

Time-series forecasting is an essential task with wide real-world applications across domains. While recent advances in deep learning have enabled time-series forecasting models with accurate predictions, there remains considerable debate over which architectures and design components, such as series decomposition or normalization, are most effective under varying conditions. Existing benchmarks primarily evaluate models at a high level, offering limited insight into why certain designs work better. To mitigate this gap, we propose TimeRecipe, a unified benchmarking framework that systematically evaluates time-series forecasting methods at the module level. TimeRecipe conducts over 10,000 experiments to assess the effectiveness of individual components across a diverse range of datasets, forecasting horizons, and task settings. Our results reveal that exhaustive exploration of the design space can yield models that outperform existing state-of-the-art methods and uncover meaningful intuitions linking specific design choices to forecasting scenarios. Furthermore, we release a practical toolkit within TimeRecipe that recommends suitable model architectures based on these empirical insights. The benchmark is available at: https://github.com/AdityaLab/TimeRecipe.

TimeRecipe: A Time-Series Forecasting Recipe via Benchmarking Module Level Effectiveness

TL;DR

TimeRecipe addresses the disconnect between holistic benchmarking and the need for understanding module-level effectiveness in time-series forecasting. It proposes a canonical, module-centered framework to evaluate pre-processing, embedding, and feed-forward components, conducting over 10,000 experiments across univariate/multivariate settings, short/long horizons, and diverse datasets. Key contributions include (i) a comprehensive modular benchmark that uncovers architectures outperforming state-of-the-art methods, (ii) correlations linking module performance to data properties, and (iii) a training-free LightGBM-based toolkit for practical architecture selection. The work demonstrates that exhaustive modular exploration yields superior performance and that design choices should align with intrinsic data characteristics, offering actionable guidance and a path toward adaptive AutoML for time-series forecasting.

Abstract

Time-series forecasting is an essential task with wide real-world applications across domains. While recent advances in deep learning have enabled time-series forecasting models with accurate predictions, there remains considerable debate over which architectures and design components, such as series decomposition or normalization, are most effective under varying conditions. Existing benchmarks primarily evaluate models at a high level, offering limited insight into why certain designs work better. To mitigate this gap, we propose TimeRecipe, a unified benchmarking framework that systematically evaluates time-series forecasting methods at the module level. TimeRecipe conducts over 10,000 experiments to assess the effectiveness of individual components across a diverse range of datasets, forecasting horizons, and task settings. Our results reveal that exhaustive exploration of the design space can yield models that outperform existing state-of-the-art methods and uncover meaningful intuitions linking specific design choices to forecasting scenarios. Furthermore, we release a practical toolkit within TimeRecipe that recommends suitable model architectures based on these empirical insights. The benchmark is available at: https://github.com/AdityaLab/TimeRecipe.

Paper Structure

This paper contains 28 sections, 6 equations, 1 figure, 30 tables, 6 algorithms.

Table of Contents

  1. Introduction
  2. TimeRecipe Framework
  3. Pre-Processing
  4. Embedding
  5. Feed-Forward Modeling
  6. TimeRecipe Benchmark
  7. Evaluation Protocol
  8. Evaluation Tasks
  9. Evaluation Dataset
  10. Evaluation Metric
  11. Results and Discussion
  12. Insights Derived from Benchmark Results
  13. TimeRecipe Yields Model Architectures Outperforming SOTA
  14. TimeRecipe Uncovers Modules Effectiveness Linking to Data Properties
  15. Practical Implications from the Insights
  16. ...and 13 more sections

Figures (1)

  • Figure 1: The proposed canonical architecture in TimeRecipe for constructing general time-series forecasting models. The canonical architecture comprises five key components: pre-processing, embedding, feed-forward modeling, projection, and post-processing.

Theorems & Definitions (3)

  • Remark 1
  • Remark 2
  • Remark 3