ARIES: Relation Assessment and Model Recommendation for Deep Time Series Forecasting

2025-09-07

AI Summary

Existing benchmarks lack diverse, interpretable temporal patterns, hindering the identification of intrinsic relationships between time-series properties and model performance; moreover, deep forecasting model selection relies on inefficient trial-and-error without interpretable guidance. To address this, we introduce the first synthetic dataset featuring explicitly controllable temporal patterns, coupled with large-scale benchmarking across 50+ models and multidimensional time-series feature analysis. We systematically quantify statistical correlations between 12 temporal attributes and model performance. Building upon these insights, we propose the first interpretable model recommendation framework for deep time-series forecasting, enabling automated, high-fidelity model selection based on input sequence characteristics. Experiments demonstrate that our framework significantly reduces model tuning costs on real-world data and improves selection accuracy by 42.6% over random choice.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models
• Planning, Routing, and Scheduling: Temporal Planning

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Recent advancements in deep learning models for time series forecasting have been significant. These models often leverage fundamental time series properties such as seasonality and non-stationarity, which may suggest an intrinsic link between model performance and data properties. However, existing benchmark datasets fail to offer diverse and well-defined temporal patterns, restricting the systematic evaluation of such connections. Additionally, there is no effective model recommendation approach, leading to high time and cost expenditures when testing different architectures across different downstream applications. For those reasons, we propose ARIES, a framework for assessing relation between time series properties and modeling strategies, and for recommending deep forcasting models for realistic time series. First, we construct a synthetic dataset with multiple distinct patterns, and design a comprehensive system to compute the properties of time series. Next, we conduct an extensive benchmarking of over 50 forecasting models, and establish the relationship between time series properties and modeling strategies. Our experimental results reveal a clear correlation. Based on these findings, we propose the first deep forecasting model recommender, capable of providing interpretable suggestions for real-world time series. In summary, ARIES is the first study to establish the relations between the properties of time series data and modeling strategies, while also implementing a model recommendation system. The code is available at: https://github.com/blisky-li/ARIES.

Problem

• Lack of diverse temporal patterns in benchmark datasets restricts systematic model evaluation.
• Absence of effective model recommendation increases time and cost for testing architectures.
• Need to establish correlation between time series properties and deep forecasting strategies.

Innovation

• Constructs synthetic dataset with distinct temporal patterns
• Establishes correlation between time series properties and models
• Implements interpretable deep forecasting model recommendation system