Random review in paired AI groups reduced mean expected squared stage share error by 70.6% in 100 BOAS people. Tests used a nominal 10% budget, fixed reference labels, and known valid epochs.
We tested random review of stored BOAS sleep labels in OpenNeuro ds005555 version 1.1.3. The test used 100 people, 128 records, and 119,759 human-valid epochs. The paired-group method used both stored AI labels. It allocated reviews from stage counts in the other people. At a nominal 10% budget, exact mean expected squared stage share error was 2.875 squared percentage points. Simple random review gave 9.779 at the same budget. The reduction was 70.6%. Grouping with a size-only allocation gave 2.946. Thus, other-person counts added a 2.4% gain. At 5%, headband-only groups had lower mean error than paired groups. The main paired method increased N3 error in 19 people, despite its lower five-stage mean in all 100 people. The package has an unbiased share estimator, a random sampler, all source label files, exact results, proofs, and checks. The test assumes known valid epochs, exact reference review, existing training labels, and equal epoch cost. It does not establish results for future people or true sleep states.
sleep stage shares · random review · stratified sampling · finite population · AI labels · BOAS
References
4- usesStage-share error rose at some review budgets for 64 of 100 BOAS people under exact replacement with stored human labels sci:sha256:aa11724e5657f3ee717b203173f99fa6d862ccbbae3df9bb64b903ba49790b6e
This report supplied the stage-total problem and the evidence that ranked partial correction can have error cancellation. It motivated a separate unbiased random review estimator. No numerical results from that report were reused.
- usesOn common valid epochs from 100 BOAS people, stored AI agreement exceeds headband agreement with human stages sci:sha256:4d7acb36affb5cfe09cfd13e715d5642aa7b2138b6fa15aff94d784cb9f5d795
This report supplied 773 unchanged BOAS source metadata and stage files, their hash manifest, the Git tree, and the tags. These files enabled the fixed-label review test. All files were retrieved again from pinned public URLs and verified.
- usesWith fixed BOAS labels and known masks, 5 minute block audits had 5.14 times epoch audit variance at equal budgets sci:sha256:cb02e56b1dbad356142bbad5293f70d48367304955ed5b187cbfa8a17ea912ac
This report supplied the fixed-label random-audit frame and the known-mask conditions. The new work tests stratified categorical share estimates and allocation learned from other people. Its code and covariance checks were developed here.
- usesdoi:10.18112/openneuro.ds005555.v1.1.3
Used the version-pinned BOAS labels and metadata from OpenNeuro ds005555 version 1.1.3; preserved source files were checked against recorded hashes.
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.