{"id":"W4403538369","doi":"10.1016/j.dajour.2024.100523","title":"A multi-criteria decision analysis framework for evaluating deep learning models in healthcare research","year":2024,"lang":"en","type":"article","venue":"Decision Analytics Journal","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Humber Polytechnic; Concordia University","funders":"","keywords":"Health care; Computer science; Management science; Decision analysis; Data science; Knowledge management; Artificial intelligence; Psychology; Political science; Engineering; Mathematics","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":[0.06642702,0.003762665,0.002835362,0.01245221,0.001764935,0.005886859,0.002636305,0.002643889,0.00319201],"category_scores_gemma":[0.08611961,0.001015035,0.003750477,0.004776467,0.002452503,0.003096988,0.004767042,0.004129724,0.0003016131],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007510052,"about_ca_system_score_gemma":0.007803164,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006362751,"about_ca_topic_score_gemma":0.006799695,"domain_scores_codex":[0.9348409,0.05134026,0.003386189,0.002073519,0.00720443,0.001154613],"domain_scores_gemma":[0.8866434,0.09498359,0.005617718,0.001713731,0.009275916,0.001765583],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.000722644,0.0005332581,0.01047191,0.00273489,0.001561618,0.000303245,0.0007193635,0.8119256,0.001617852,0.06144955,0.003346219,0.1046139],"study_design_scores_gemma":[0.0001171608,0.0006727016,0.001275552,0.0006111836,0.0002462632,0.00006646434,0.0003800619,0.9282437,0.00115288,0.06537266,0.001771728,0.00008963247],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.04534753,0.003561839,0.9379607,0.003643433,0.0001408572,0.002486868,0.0009110401,0.0003154117,0.005632212],"genre_scores_gemma":[0.4067746,0.001098783,0.5870103,0.0005984723,0.0001447715,0.002960659,0.0007012543,0.00005638055,0.0006548194],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.06642702,"threshold_uncertainty_score":0.3513039,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2845120089915341,"score_gpt":0.5449850039244639,"score_spread":0.2604729949329298,"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."}}