{"id":"W4307948995","doi":"10.3390/forecast4040048","title":"Precision and Reliability of Forecasts Performance Metrics","year":2022,"lang":"en","type":"article","venue":"Forecasting","topic":"Forecasting Techniques and Applications","field":"Decision Sciences","cited_by":23,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal; Polytechnique Montréal","funders":"Mitacs","keywords":"Variance (accounting); Reliability (semiconductor); Metric (unit); Computer science; Sensitivity (control systems); Noise (video); Series (stratigraphy); Selection (genetic algorithm); Econometrics; Quality (philosophy); Model selection; Performance metric; Statistics; Data mining; Reliability engineering; Machine learning; Artificial intelligence; Mathematics; Engineering","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.03374534,0.001828279,0.001868589,0.00852739,0.0006637075,0.003569803,0.001061572,0.001805434,0.001518258],"category_scores_gemma":[0.1601459,0.0004935236,0.001302557,0.004187729,0.001122564,0.003451384,0.001884259,0.001139826,0.0007944987],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001288042,"about_ca_system_score_gemma":0.00109484,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003998267,"about_ca_topic_score_gemma":0.001621644,"domain_scores_codex":[0.9598054,0.01069565,0.004700711,0.005186008,0.0181879,0.001424332],"domain_scores_gemma":[0.8387008,0.09796711,0.01676376,0.01760744,0.02795425,0.001006595],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001600994,0.000206138,0.1490568,0.001202515,0.001441865,0.0004269974,0.001004292,0.5305982,0.01050724,0.01078721,0.005742875,0.287425],"study_design_scores_gemma":[0.0001146868,0.001509187,0.1618456,0.0006000072,0.0005861589,0.0009966897,0.0009256139,0.762314,0.03136682,0.02507513,0.01408793,0.0005781996],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5151539,0.00656492,0.4472659,0.001457737,0.0007997042,0.0005286289,0.005799403,0.00425736,0.01817238],"genre_scores_gemma":[0.9671377,0.000569067,0.02918586,0.00007027223,0.0002122125,0.0001107398,0.001754193,0.0001952587,0.0007645713],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9662547,"threshold_uncertainty_score":0.1784646,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1467905736224284,"score_gpt":0.3568156231551683,"score_spread":0.2100250495327398,"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."}}