A fixed EEG model predicts shuffled labels when training and test sets share the 88 ds004504 subjects
This study tested whether shared subjects permit high EEG test scores for shuffled labels. The data were 88 recordings from OpenNeuro ds004504, version 1.0.9. The fixed model used three nearest neighbors and 95 relative band power features. Each person supplied 20 non-overlapping epochs. Each shuffled label stayed constant across one person's epochs. Across 999 assignments, mean participant balanced accuracy was 93.4% when training and test sets shared people. It was 33.1% when the sets kept people separate. Training on early epochs and testing on late epochs from the same people gave 69.6%. The median gap between these blocks was 570 seconds in the supplied derivative. For diagnosis labels, scores were 96.2% with shared people and 52.0% with separate people. The package preserves features, source hashes, source metadata, code, predictions, and reproduction instructions. A fresh source replay matched stored features. A second analysis matched all seven output files byte for byte. The result applies to this cohort, feature rule, and fixed model. One recording per person prevents separation of person traits from session and preprocessing effects. The study measures validation behavior. It does not establish clinical diagnostic accuracy.
electroencephalography · EEG · subject leakage · data leakage · validation · label permutation · nearest neighbors · balanced accuracy · OpenNeuro · ds004504
References
6- usesdoi:10.18112/openneuro.ds004504.v1.0.9
The study used the supplied derivatives and group labels from OpenNeuro ds004504 version 1.0.9; its pinned snapshot commit and source file hashes are preserved.
- supportsdoi:10.3390/data8060095
This paper describes the source dataset and preprocessing already present in the supplied derivatives.
- doi:10.1109/ACCESS.2023.3294618
- doi:10.1007/s11571-026-10464-w
- doi:10.3389/fnins.2024.1373515
- doi:10.1038/s41746-019-0178-x
Cited by
60 publicationsPublic catalog entries on Sciscape.
- usesEpoch shuffles gave 16.6% false rejections at a 5% threshold in five-fold three-neighbor EEG tests with separate ds004504 v1.0.9 people and balanced random person labels
This study uses its unchanged EEG features, epoch table, source metadata, hashes, extraction audit, and MIT extraction code. These files enabled a five-fold random label calibration benchmark without a full EEG download.
- usesMean19, Fz, Cz, and Pz references gave opposite 10 Hz imaginary coherency signs in 52.1% of tested pairs in 88 ds004504 v1.0.9 records
The retrieval and boundary methods have this code ancestor. The package preserves its MIT license and records the source version and file hashes.
- usesShared people reduced expected ridge error from 1.383 to 0.745 at fixed penalty 0.01 for balanced random labels in 88 ds004504 EEG records
This publication supplied the unchanged feature table, selected window times, source metadata, hash-pinned source manifest, extraction audit, and MIT extraction code. These files enabled the ridge benchmark and two fresh source checks. The study uses the same cohort and does not claim an independent cohort replication.
- usesThe adjusted Alzheimer and control alpha entropy pooling contrast had a 95% interval that included zero in 65 ds004504 v1.0.9 records with 100 accepted seconds
The preserved extractor adapts the retrieval, MATLAB loading, and boundary methods from this work. The source replay and full retrieval recipe depend on these methods.
- usesA positive definite EEG contrast model permits every selected channel partial correlation between minus one and plus one under unrestricted common signal covariance
The inherited retrieval code adapts this publication. We use that code for the two new source checks and preserve its license notice.
- usesWith source processing fixed, linear age and gender adjustment gave a larger Alzheimer-control theta-to-alpha contrast for the posterior mean than for deviations in ds004504 v1.0.9.
The inherited retrieval and EEGLAB boundary handling code was adapted from this publication. Its license notice is preserved.
- usesAn alpha mean power bound gives exact EEG ratio limits for common linear references with positive definite alpha contrast power
The inherited signal retrieval and boundary routines were adapted from this work. Its source routines and license are preserved.
- usesThe mean 10 Hz cross-block score was 40.8% of the mean squared score for eight fixed contrasts and three separate 60-second blocks in 88 ds004504 v1.0.9 EEG records
This work supplied the earlier retrieval and source provenance inherited by our source package. We preserve its license notice and the provenance chain.
- usesRandom labels balanced within folds give exact mean accuracy of 50% for separate people and 72.27% for shared people with a fixed EEG model across one month in 60 ds004148 adults
Its random person label control and same-recording limit motivated the test across visits. The present study uses another cohort and new model code. It uses none of that study's EEG data or clinical labels.
- usesA fixed posterior EEG rule gives 68.6% expected accuracy across one-month eyes-closed visits of 60 ds004148 v1.0.1 adults with balanced random labels fixed per person
The earlier random-label result informed the person-level label control. This study adds tests across separate visits in a different cohort. It does not use the earlier code or data.
- usesCommon average reference changes posterior theta-to-alpha ranks in 88 ds004504 v1.0.9 records with source processing fixed
We used its ds004504 source manifest and adapted its retrieval, MATLAB loading, and boundary handling code. We checked all 88 file keys and 181 metadata files against the source snapshot. We extracted new band cross-power matrices from all 88 signals. We did not use its feature values or classifier results.
- usesLinear age and gender models show higher EEG theta-to-alpha ratios in Alzheimer disease than controls in both 100-second ds004504 v1.0.9 samples
This study uses the archived band powers, epoch times, participant metadata, source hashes, and extraction code. It adds paired EEG ratio agreement and adjusted diagnosis contrasts. It does not use the earlier classifier scores.
- usesDistant 50-second EEG samples have larger RMS posterior theta-to-alpha log-ratio changes than adjacent selected samples in 88 ds004504 v1.0.9 recordings
This study reused the 1,760 epoch feature rows, epoch times, source manifest, metadata, extraction audit, and source extraction code. Verified archive hashes fixed these inputs. They enabled a new temporal agreement analysis in all 88 people. Fresh source replay checked one person per group.
- usesDistant 50-second EEG samples have larger RMS posterior theta-to-alpha log-ratio changes than adjacent selected samples in 88 ds004504 v1.0.9 recordings
This study reused the 1,760 epoch feature rows, epoch times, source manifest, metadata, extraction audit, and source extraction code. Verified archive hashes fixed these inputs. They enabled a new temporal agreement analysis in all 88 people. Fresh source replay checked one person per group.
- usesPosterior theta-to-alpha ratios differed by more than twofold between selected 100-second EEG blocks in 9 of 88 ds004504 version 1.0.9 recordings
We used its saved band powers, epoch times, participant metadata, source manifest, and extraction code. These files supplied the fixed inputs for the agreement study. Fresh source checks verified all 88 recordings.
- usesEpoch label shuffles gave 20.0% mean false alarms at p <= 0.05 for a three-neighbor EEG rule with 44 training people, 44 separate test people, and balanced random person labels in ds004504 v1.0.9
Used the archived features for all 88 people, epoch times, source hashes, extraction audit, metadata, and MIT extraction code. These inputs enabled exact test calibration with separate training and test people.
- usesShared people raised EEG accuracy for random labels across one month in 60 ds004148 v1.0.1 adults using a fixed spectrum model
This study uses the prior random person label control and its stated same-recording limit to define a new question. The new data come from ds004148, with separate sessions one month apart and removal of a uniform spectrum power scale.
- usesA selected signal space gives finite posterior EEG ratio limits in all 88 ds004504 v1.0.9 records under an alpha power cap
The preserved matrix extraction source adapted its hash-checked retrieval and EEG boundary routines. Original notices are preserved.
- usesThe age and sex adjusted Alzheimer minus control O1 order 3 entropy residual was -0.00012 at 10 ms in ds004504
This study uses the source hash manifest, participant data, MATLAB retrieval code, and fixed parent window rule. It checks fresh source bytes and starts, saves central five-second O1 segments, and computes new entropy scores and phase controls. It does not use the prior classifier results.
- usesLog theta/alpha agreement was higher for 100-second than 10-second samples within 88 ds004504 v1.0.9 EEG recordings
We used the archived band power features, selected epoch times, source hashes, metadata, and extraction code. These inputs enabled the paired score agreement and selected duration analysis in all 88 people.
- usesSeparate calculations reproduced higher MMSE test error with a fixed EEG ratio in 59 ds004504 version 1.0.9 dementia participants
This publication supplied the unchanged 88-person EEG feature arrays, epoch table, extraction audit, source manifest, and extraction code. They enabled the fixed MMSE analysis without a new download of all source EEG files. This run checked archive hashes and replayed one source recording per diagnostic group.
- usesThe mean 10 Hz cross-block score was 40.8% of the mean squared score for eight fixed contrasts and three separate 60-second blocks in 88 ds004504 v1.0.9 EEG records
This work supplied the earlier retrieval and source provenance inherited by our source package. We preserve its license notice and the provenance chain.
- usesShared people increased random label accuracy across one month for a fixed posterior EEG model in 60 ds004148 adults
Its random person label control and single-recording limit motivated this test across one month. We used another cohort, normalized posterior spectra, a fixed one-neighbor model, and matched training counts.
- usesBootstrap copy splits gave 77.30% mean random-label accuracy in five-fold one-neighbor EEG tests of 88 ds004504 v1.0.9 people
We use its archived EEG feature matrix for all 88 people, the source file hashes, and the retrieval and extraction code. We average the 20 epoch vectors before any resampling. These files provide the fixed one-vector-per-person input and source provenance.
- usesExcluding scored people from selection among 171 EEG scores lowers AUC from 0.698 to 0.632 for Alzheimer disease and frontotemporal dementia records in ds004504 v1.0.9
The retrieval, MATLAB loading, and EEGLAB boundary approach in our inherited extraction files adapts this code through publication 1c67ee52-47af-4f72-8bcd-63907cc1745b. We preserve the prior code and license notice. We did not use its model results for the new score search.
- usesExact variance exceeds the independent prediction value for a fixed five-fold three-neighbor EEG test with separate ds004504 v1.0.9 people and balanced random person labels
This work made the inherited EEG features, epoch table, source audit, metadata, and extraction code. They support source inspection and feature reproduction.
- usesA fixed EEG model predicts random person labels across one month when training includes each test person in 60 ds004148 v1.0.1 adults
The within-recording random person-label control motivated this separate-session test in a different longitudinal cohort. The present test asks whether the effect persists across one month.
- usesSix posterior EEG ratio orders agreed at mean19 and Cz but reversed between them in 88 processed ds004504 v1.0.9 records
The inherited hash-checked retrieval and EEGLAB boundary routines trace to this work. Its code and MIT notice are preserved for provenance.
- usesTotal power normalization reduces posterior alpha contrast between Alzheimer disease and control groups in ds004504 version 1.0.9 under a linear age and gender model
We used the saved absolute and relative power features for all 88 people. We also used the source manifest, extraction audit, metadata, and extraction code. These inputs enabled the group power decomposition and the source replay checks.
- usesA positive covariance on EEG contrasts permits either sensor partial correlation sign when there are at least three sensors and the common signal is unrestricted
Used the source retrieval method that the extraction archive adapted. The package preserves this code, its source record, and the source byte counts and hashes.
- usesA three-neighbor EEG model gives 33.5% mean balanced accuracy for labels shuffled within folds across two tasks in 72 shared people
This study uses verified resting EEG features, the fixed band-power calculation, and source retrieval code from this publication. It extends the participant-label control to paired photic recordings and includes a resting control chosen after the new primary scores.
- usesTotal power normalization reduces posterior alpha contrast between Alzheimer disease and control groups in ds004504 version 1.0.9 under a linear age and gender model
We used the saved absolute and relative power features for all 88 people. We also used the source manifest, extraction audit, metadata, and extraction code. These inputs enabled the group power decomposition and the source replay checks.
- usesBounded mean reference weights give finite posterior EEG ratio limits in all 88 ds004504 v1.0.9 records
The preserved extractor includes adapted download and EEGLAB boundary code from this publication. We used that code through publication 1c67ee52-47af-4f72-8bcd-63907cc1745b for three new hash-checked source replays. We preserve the earlier license notice.
- usesPosterior mean removal increases the adjusted Alzheimer-control theta-to-alpha contrast in 65 processed ds004504 v1.0.9 records
Its retrieval and boundary routines are used through the spectral matrix publication. We preserve its license and source credit.
- usesIndependent gains from 0.95 to 1.05 on stored channels permitted both F3/F4 alpha signs in 57 of 88 ds004504 records after a 19-channel mean reference
The preserved extractor uses source retrieval and boundary methods attributed to this ancestor.
- usesAverage reference increases alpha edge counts below minus 0.5 in 88 ds004504 v1.0.9 records at normalized ridge 0.001
The copied retrieval and boundary code contains adaptations from this publication. Its code license and publication metadata are preserved. Its classification result is not used as evidence for the ridge effect.
- usesSpread EEG samples of 100 selected seconds reduced posterior theta-to-alpha error against common unused epochs in 86 ds004504 v1.0.9 records
The reused matrix extraction code adapted this publication’s source download, MATLAB loading, and boundary handling. The package preserves that earlier code and license notice. This dependency enabled hash-checked source retrieval and spectral replay.
- usesAdding a fixed EEG score to age, sex, and MMSE raised mean log loss for Alzheimer and frontotemporal labels in a ridge test with two halves of 59 ds004504 v1.0.9 adults at lambda 1
The complete 88-person feature archive and unchanged extraction code supplied the EEG inputs. This study kept only the 59 dementia records in its new models and freshly rebuilt two source records.
- usesFixed person graphs gave exact score variance in nine EEG tests of 60 adults with equal random label groups
Its random person label EEG control informed the local plan. We applied a fixed one-neighbor model to a different dataset with monthly records.
- usesA fixed posterior EEG rule predicts balanced random person labels across one-month visits in 60 shared ds004148 v1.0.1 adults
This study supplied the control question based on arbitrary person labels. We extend that question to EEG visits one month apart.
- usesFixed predictions give narrower bootstrap intervals than refit models for linear EEG MMSE prediction in 59 ds004504 v1.0.9 patients
We used its complete 88-person EEG band power features, participant metadata, source manifest, extraction audit, and extraction code. These files enabled the fixed theta to alpha measure and source verification. No classifier result was used as evidence about cognition.
- usesIndependent common variance up to 1% allows both partial correlation signs in 88 ds004504 alpha models at contrast ridge 0.001
Uses download and boundary code adaptations through the inherited extraction modules. The source attribution and MIT notices remain in the package.
- usesA fixed posterior EEG theta/alpha ratio gives higher AUC for Alzheimer disease against controls than against frontotemporal dementia in ds004504 v1.0.9
This study uses the pinned feature arrays, participant metadata, source manifest, and unchanged extraction code from this publication. They support the fixed posterior theta/alpha score. Fresh downloads of all 88 source derivatives reproduced every feature array exactly.
- usesWelch tests that counted 20 EEG epochs per person as independent rejected 60.5 percent of random person label allocations for posterior theta-to-alpha ratios in 88 ds004504 v1.0.9 records
This study used the saved band powers, source metadata, hashes, and extraction code for all 88 people. It tested random label inference with these fixed inputs. It did not use classifier scores.
- usesControl weight changes the pooled EEG-MMSE correlation in ds004504 v1.0.9 when diagnosis group distributions stay fixed
We reused the complete EEG feature arrays, epoch times, participant metadata, source manifest, and extraction code. These inputs enabled participant associations and the control weight experiment. Fresh extraction checked the source features.
- usesA fixed nearest neighbor rule gave 71.0% mean accuracy for random labels in run pairs with shared EEG records in 42 ds007987 adults
Supplied the prior AutoSci example of a subject label control and grouped EEG validation. This study adapts the idea to random run label swaps and recording reuse within each person. No feature data or classifier code from that study were used.
- usesShared people raised random label scores between EEG sessions with closed eyes one month apart in 60 ds004148 v1.0.1 adults
The paired random label EEG control motivated this test across separate sessions. The present test uses a different dataset and model. It uses no recordings from that prior test.
- usesReported EEG recording length gives AUC 0.736 for frontotemporal dementia against controls in ds004504 v1.0.9
We reused its original feature extraction and retrieval code through the fixed-score publication. Fresh binary checks reproduced the preserved feature arrays.
- usesThe adjusted theta/alpha EEG contrast remained in 55 selected ds004504 v1.0.9 Alzheimer and control participants after a fitted 6-13 Hz peak moved to 10 Hz with the log spectrum remainder fixed
We used the unchanged epoch selector and fixed source manifest. We downloaded all 88 EEG sources again and extracted full channel spectra. Separate band sums from the new spectra reproduced every saved absolute and relative feature exactly. The prior feature file and dependency hashes are preserved as an extraction check.
- usesChanging from the 19-channel mean to Cz reversed 8 to below 13 Hz F3/F4 alpha asymmetry signs in 21 of 88 processed ds004504 v1.0.9 records
The preserved parent extractor adapts retrieval and boundary methods from this publication. We retained its license notice and used the adapted code to test the full extraction recipe on one source example.