Research map / Time series / Causal and generative time series
Beyond Model Ranking: Predictability-Aligned Evaluation for Time Series Forecasting
Wanjin Feng, Yuan Yuan, Jingtao Ding, Yong Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| vary Welch window-length fraction from 0.25 to 0.35 (ablation) | SCP · Weather (N=96) | SCP mean: 0.344 -> 0.367 | No clear effect |
| vary overlap ratio 0.45 to 0.55 (ablation) | SCP · Weather (N=96) | SCP mean: 0.362 / 0.344 / 0.351 | No clear effect |
| swap taper window Hann to Hamming or Blackman (ablation) | SCP · Weather (N=96) | SCP mean: 0.345 -> 0.351 / 0.335 | No clear effect |
| vary history length 96 to 720 (ablation) | iTransformer · ETTh1 (Ny=336) | Pearson R with MSElb: 0.808, 0.842, 0.899, 0.832 | No clear effect |
| longer history 96 to 336 (ablation) | DLinear · ETTh1 (Ny=336) | MSE: 0.491 -> 0.447 | Improved |
| longer history 336 to 720 (ablation) | iTransformer · ETTh1 (Ny=336) | MSE: 0.471 -> 0.480 | Hurt |
| stratify by SCP predictability: nonlinear models in low-P bins | TimesNet vs DLinear · ETTh1 (N=96) | MSE per P bin | Improved |
| linear model in high-P regime | DLinear · ETTh1 (N=96) | MSE per P bin | Improved |
| higher-frequency bands LUR: DLinear vs Transformers | DLinear · ETTh1 band-wise | LUR | Improved |
Closest papers by meaning
- Introducing Spectral Attention for Long-Range Dependency in Time Series Forecasting (NeurIPS 2024 · similarity 0.86)
- fev-bench: A Realistic Benchmark for Time Series Forecasting (NeurIPS 2026 · similarity 0.86)
- TimeBase: The Power of Minimalism in Efficient Long-term Time Series Forecasting (ICML 2025 · similarity 0.86)
- ProbTS: Benchmarking Point and Distributional Forecasting across Diverse Prediction Horizons (NeurIPS 2024 · similarity 0.85)
- Improving Time Series Forecasting via Instance-aware Post-hoc Revision (NeurIPS 2025 · similarity 0.85)
- Less is More: Unlocking Specialization of Time Series Foundation Models via Structured Pruning (NeurIPS 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
