New cross-validation method boosts time series forecasting accuracy

A new paper proposes a series-wise cross-validation method for global time series forecasting models, particularly those based on deep learning. This approach partitions data across both series and time, simulating the emergence of new series not seen during training. Benchmarking various instantiations like K-fold CV and bootstrapping on multiple datasets and neural network architectures, the study found that series-wise CV improves downstream forecasting accuracy by guiding model selection towards configurations that generalize well across both time and individual series. AI

IMPACT Introduces a novel cross-validation technique that could improve the generalization and accuracy of global time series forecasting models, particularly those leveraging deep learning.

RANK_REASON The item is an academic paper published on arXiv detailing a new methodology for time series forecasting. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv stat.ML →

• alphaXiv
• arXiv
• CatalyzeX Code Finder for Papers
• computer vision
• Connected Papers
• CORE Recommender
• DagsHub
• Gotit.pub
• Hugging Face
• K-fold CV
• Litmaps
• Monte Carlo CV
• ScienceCast
• scite Smart Citations

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →