Prediction Without Generalization: A Reanalysis of EEG Dream Recall Models in the DREAM Database
Zenodo (CERN European Organization for Nuclear Research) June 8, 2026 DOI: 10.5281/zenodo.20597471 (opens in new tab) via OpenAlex
Summary
AI-generated from the abstractAcross 6 datasets, 148 subjects, and 1,267 awakenings, 52 validation analyses found no evidence of above-chance generalization after bootstrap testing. Dataset identity accounted for 74.5% of predictive variance, indicating that models did not transfer across datasets. All code and data are publicly available.
Study at a glance
| Characteristics | Validation analyses Peer reviewed |
|---|---|
| Sample size | 148 |
| Population | Subjects in sleep and wakefulness research |
| Keywords | Recall Generalization Dream Variance accounting Code set theory |
| Key finding | No dataset-stage combination showed evidence of above-chance generalization after bootstrap testing, and dataset identity dominated predictive variance (74.5%). |
Abstract
52 validation analyses across 6 datasets, 148 subjects, and 1,267 awakenings. No dataset-stage combination showed evidence of above-chance generalization after bootstrap testing. Dataset identity dominates predictive variance (74.5%). All code and data publicly available.