Reproduce “Are We Really Benchmarking Forecasting Models? The Impact of Preprocessing on Time Series Performance” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Guilherme Afonso Galindo Padilha, Paulo S. G. de Mattos Neto, Rafael M. O. Cruz
The authors test how preprocessing affects forecasting performance across about 29,000 univariate M4 series using 11 models and 16 reversible pipelines. They compare fixed preprocessing with an oracle that selects the best pipeline separately for each series. Informer’s mean MASE falls from 106.75 to 1.14, a reported gain of 86.73%. The oracle is retrospective; hyperparameters are largely fixed.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.