{"id":"W4309636459","doi":"10.1007/s11634-022-00528-0","title":"A power-controlled reliability assessment for multi-class probabilistic classifiers","year":2022,"lang":"en","type":"article","venue":"Advances in Data Analysis and Classification","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":4,"is_retracted":false,"has_abstract":false,"ca_institutions":"Western University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Probabilistic logic; Reliability (semiconductor); Computer science; Probabilistic classification; Classifier (UML); Statistic; Bayesian probability; Artificial intelligence; Machine learning; Class (philosophy); Naive Bayes classifier; Sample size determination; Sampling distribution; Data mining; Mathematics; Statistics; Power (physics); Support vector machine","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01330647,0.001408744,0.001591396,0.003121646,0.0009901348,0.002389849,0.002815577,0.001990469,0.002686918],"category_scores_gemma":[0.0635392,0.0005953461,0.001420805,0.001827172,0.002360127,0.003794708,0.002420679,0.00205662,0.0008435481],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001083029,"about_ca_system_score_gemma":0.001174999,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001422423,"about_ca_topic_score_gemma":0.001043025,"domain_scores_codex":[0.9891439,0.004048062,0.0004643455,0.001761784,0.004250381,0.0003315706],"domain_scores_gemma":[0.9593536,0.02275105,0.002324846,0.006704351,0.008526643,0.0003394915],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0009073204,0.0002416494,0.007044361,0.0006892686,0.0003633817,0.0003274956,0.0007265987,0.3947591,0.03790169,0.097262,0.004568814,0.4552083],"study_design_scores_gemma":[0.00002026836,0.0002661149,0.002073162,0.00005443479,0.00008198764,0.000221744,0.00004406545,0.9502766,0.008158584,0.03725028,0.001505589,0.00004726527],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.009315494,0.0002201483,0.988844,0.00008139711,0.00003336987,0.00007397034,0.00004911564,0.0002622081,0.001120269],"genre_scores_gemma":[0.6957506,0.0004283348,0.2997295,0.0001732647,0.0002568743,0.0004879978,0.0004190274,0.0003734051,0.002381019],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01330647,"threshold_uncertainty_score":0.07037216,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1683283872417896,"score_gpt":0.4828551957848841,"score_spread":0.3145268085430944,"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."}}