{"id":"W4407626533","doi":"10.1016/j.mlwa.2025.100630","title":"Predicting classification errors using NLP-based machine learning algorithms and expert opinions","year":2025,"lang":"en","type":"article","venue":"Machine Learning with Applications","topic":"Imbalanced Data Classification Techniques","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University; Western University","funders":"","keywords":"Artificial intelligence; Computer science; Machine learning; Natural language processing","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01422957,0.001461086,0.001266237,0.003834234,0.0006747578,0.002660723,0.001346887,0.002604725,0.001888866],"category_scores_gemma":[0.06073552,0.0002995355,0.000731072,0.001728015,0.000912472,0.003560605,0.001236638,0.002068991,0.001534632],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001077134,"about_ca_system_score_gemma":0.0009926093,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00291963,"about_ca_topic_score_gemma":0.00290916,"domain_scores_codex":[0.9921143,0.003481926,0.0007672846,0.001839821,0.00142729,0.0003693284],"domain_scores_gemma":[0.929205,0.05667561,0.004197573,0.00352462,0.005727679,0.0006694458],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001337532,0.0008732218,0.08395175,0.0004330944,0.000409041,0.000393751,0.0009901271,0.275389,0.00698697,0.004482664,0.008800643,0.6159521],"study_design_scores_gemma":[0.00002443744,0.0001486253,0.007083834,0.00005272858,0.0000302625,0.00008285684,0.0001916487,0.9806193,0.00413635,0.006770878,0.0008291064,0.00002991404],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.367391,0.001081397,0.6224897,0.001276888,0.0002740989,0.0002192298,0.0006764836,0.002012412,0.004578946],"genre_scores_gemma":[0.8796495,0.0002034737,0.1165754,0.0002871522,0.0002013169,0.0001307839,0.001170748,0.0001230886,0.001658545],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01422957,"threshold_uncertainty_score":0.07525402,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02403382708901836,"score_gpt":0.3024235520663833,"score_spread":0.278389724977365,"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."}}