{"id":"W2160736363","doi":"10.1016/j.nicl.2014.04.004","title":"Biased binomial assessment of cross-validated estimation of classification accuracies illustrated in diagnosis predictions","year":2014,"lang":"en","type":"article","venue":"NeuroImage Clinical","topic":"EEG and Brain-Computer Interfaces","field":"Neuroscience","cited_by":125,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University","funders":"European Regional Development Fund; James S. McDonnell Foundation","keywords":"Permutation (music); Binomial test; Binomial distribution; Cross-validation; Random forest; Multivariate statistics; Neuroimaging; Continuity correction; Statistics; Artificial intelligence; Negative binomial distribution; Computer science; Pattern recognition (psychology); Medicine; Beta-binomial distribution; Mathematics; Psychiatry","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0008562147,0.0001372875,0.0003438668,0.0001707824,0.00005722233,0.00006970541,0.0003411644,0.0001316736,0.00004310632],"category_scores_gemma":[0.005706856,0.0001269001,0.0001182294,0.0004591608,0.0005452121,0.0003434532,0.00008164393,0.0003333378,0.000007656574],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001948154,"about_ca_system_score_gemma":0.00008350024,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001754453,"about_ca_topic_score_gemma":0.000005564162,"domain_scores_codex":[0.9971972,0.0005976076,0.001234506,0.0004989045,0.0002917096,0.0001801477],"domain_scores_gemma":[0.9954544,0.00341198,0.0005272867,0.0004147431,0.0001246671,0.00006690731],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","study_design_scores_codex":[0.0001773106,0.0023875,0.461282,0.0001198487,0.00001346932,0.000009424768,0.0001544936,0.03472199,0.4845453,0.001544213,0.0007852993,0.0142591],"study_design_scores_gemma":[0.0006966207,0.0003521555,0.5261099,0.00003870466,0.00001071032,0.000002168557,0.000005641332,0.3794567,0.0930893,0.00007089647,0.00009752728,0.00006961065],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9914183,0.000001789049,0.006572693,0.000403175,0.0005658834,0.000281404,0.00006121574,0.00007374118,0.0006217802],"genre_scores_gemma":[0.9986197,0.00002503644,0.001044807,0.0001638312,0.00006176061,0.00002724674,0.00001761526,0.00001358701,0.00002642732],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.391456,"threshold_uncertainty_score":0.6832052,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1874704572399876,"score_gpt":0.4606089384883991,"score_spread":0.2731384812484116,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}