M4

The M4 dataset is a comprehensive benchmark containing 100,000 time series from various domains, used to evaluate the performance of forecasting methods.

Papers using M4 (15)

One-for-All: A Lightweight Stabilized and Parameter-Efficient Pre-trained LLM for Time Series Forecasting2026Forking-Sequences2025Optimal starting point for time series forecasting2024N-BEATS: Neural basis expansion analysis for interpretable time series forecasting2019 · 517 citesDeep Learning for Time Series Forecasting: Tutorial and Literature Survey2020 · 345 citesForecasting with time series imaging2019 · 102 citesTopological Attention for Time Series Forecasting2021 · 15 citesPerformance of Deep Learning models with transfer learning for multiple-step-ahead forecasts in monthly time series2022 · 11 citesCorrelated daily time series and forecasting in the M4 competition2020 · 8 citesfETSmcs: Feature-based ETS model component selection2022 · 6 citesPersistence Initialization: A novel adaptation of the Transformer architecture for Time Series Forecasting2022 · 1 citesImproving forecasting by subsampling seasonal time series2021Pastprop-RNN: improved predictions of the future by correcting the past2021IDEA: Interpretable Dynamic Ensemble Architecture for Time Series Prediction2022Variational Mode Decomposition and Linear Embeddings are What You Need For Time-Series Forecasting2024