{"id":"W4251121144","doi":"10.31234/osf.io/b2sq7","title":"Machine Learning to Analyze Single-Case Graphs: A Comparison to Visual Inspection","year":2021,"lang":"en","type":"preprint","venue":"","topic":"Behavioral and Psychological Studies","field":"Psychology","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"St. Lawrence College; Université de Montréal; Institut universitaire en santé mentale de Montréal; Institut Universitaire en Santé Mentale de Québec","funders":"","keywords":"Visual inspection; Computer science; Machine learning; Reliability (semiconductor); Artificial intelligence; Contrast (vision); Graph; Word error rate; Learning effect; Power (physics); Theoretical computer science","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":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0002826484,0.0004731145,0.0008173743,0.0003146668,0.0002968843,0.0001651741,0.0002355584,0.0004657433,0.004972345],"category_scores_gemma":[0.0000680836,0.0003944184,0.0003129323,0.0007413721,0.00005345931,0.00002763167,0.0008798568,0.001379979,0.0007232167],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009953327,"about_ca_system_score_gemma":0.000008554702,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007006449,"about_ca_topic_score_gemma":0.003960851,"domain_scores_codex":[0.9969845,0.0003703482,0.0006317411,0.001259695,0.0002486309,0.0005050424],"domain_scores_gemma":[0.9988257,0.00008343616,0.0001652387,0.0004342679,0.0001576945,0.0003336721],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001160684,0.008639554,0.4990647,0.00009859953,0.00125725,0.006511528,0.03055533,0.002005053,0.001849871,0.001237021,0.03800653,0.4096139],"study_design_scores_gemma":[0.003315969,0.01151508,0.7562834,0.0006300383,0.001360321,0.002315857,0.05670009,0.001264252,0.0005505072,0.001072809,0.1587682,0.006223454],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9622774,0.001080335,0.006338067,0.001296982,0.002717152,0.0005937298,0.00002028067,0.0006338969,0.02504214],"genre_scores_gemma":[0.992984,0.00001656048,0.001375071,0.00115903,0.00023449,0.0002221286,0.00009381032,0.00003755447,0.003877363],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4033904,"threshold_uncertainty_score":0.9998507,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2933103595014251,"score_gpt":0.4171803661840111,"score_spread":0.123870006682586,"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."}}