{"id":"W4366779041","doi":"10.1016/j.neuroimage.2023.120115","title":"Relationship between prediction accuracy and feature importance reliability: An empirical and theoretical study","year":2023,"lang":"en","type":"article","venue":"NeuroImage","topic":"Functional Brain Connectivity Studies","field":"Neuroscience","cited_by":62,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University; Mila - Quebec Artificial Intelligence Institute; Montreal Neurological Institute and Hospital","funders":"National Institutes of Health; National Institute on Drug Abuse; National Supercomputing Centre Singapore; National Institute of Mental Health; Medical Research Council; National Medical Research Council; National Research Foundation Singapore; National Research Foundation","keywords":"Feature (linguistics); Reliability (semiconductor); Univariate; Feature selection; Artificial intelligence; Computer science; Pattern recognition (psychology); Predictive modelling; Correlation; Statistics; Machine learning; Data mining; Mathematics; Multivariate statistics","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0005995778,0.0001591428,0.0001796354,0.0001029073,0.0004604585,0.0001015065,0.0001083795,0.00007100405,0.000008189936],"category_scores_gemma":[0.03842486,0.0001410463,0.00002538523,0.0005364534,0.0004551615,0.0004707965,0.0002133333,0.0004317288,0.00001883253],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001985101,"about_ca_system_score_gemma":0.00001993963,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000002100286,"about_ca_topic_score_gemma":0.000004102745,"domain_scores_codex":[0.9980068,0.0003757807,0.0001817607,0.0008644643,0.0003417625,0.000229463],"domain_scores_gemma":[0.9853582,0.01406334,0.00005248794,0.0003739697,0.00003419171,0.0001177758],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00003664201,0.00006983158,0.9931626,0.00001239623,0.000001959094,0.00003688856,0.0004834271,0.000002987222,0.001777872,0.001630635,0.00250204,0.0002827474],"study_design_scores_gemma":[0.0003580475,0.000468814,0.9860048,0.000003158654,0.00002283803,0.00002022728,0.0001933463,0.000591835,0.0001908861,0.01135731,0.0006781673,0.0001105826],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9851276,0.00001052285,0.00002192923,0.0134758,0.0001450458,0.0004899348,0.00005895067,0.0003772276,0.0002929867],"genre_scores_gemma":[0.998638,0.000009343299,0.00004359233,0.0009653004,0.0001505794,0.00003631564,0.000006287988,0.00002268117,0.000127918],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03782529,"threshold_uncertainty_score":0.9696749,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09338867884370698,"score_gpt":0.3566800502773856,"score_spread":0.2632913714336786,"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."}}