{"id":"W4388219066","doi":"10.21203/rs.3.rs-3516908/v1","title":"Long-term evaluation of machine learning based methods for air emission monitoring","year":2023,"lang":"en","type":"preprint","venue":"Research Square","topic":"Air Quality Monitoring and Forecasting","field":"Environmental Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Cenovus Energy (Canada); University of Calgary","funders":"","keywords":"Overfitting; Mean squared error; Machine learning; Artificial neural network; Computer science; Term (time); Artificial intelligence; Audit; Boosting (machine learning); Algorithm; Gradient boosting; Statistics; Mathematics; Random forest; Accounting","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":[],"consensus_categories":[],"category_scores_codex":[0.02533587,0.0002281432,0.000337055,0.000243233,0.0004435458,0.00005238543,0.0004978665,0.0003149011,0.0001823618],"category_scores_gemma":[0.005888528,0.0002195665,0.0002080867,0.0004435251,0.000159535,0.00007954388,0.001364522,0.001332461,0.00003388714],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009064205,"about_ca_system_score_gemma":0.0001475067,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006460874,"about_ca_topic_score_gemma":0.000008962696,"domain_scores_codex":[0.9934738,0.002573154,0.0004732492,0.0007246811,0.002156393,0.0005987085],"domain_scores_gemma":[0.9967549,0.001952592,0.0002484942,0.0005619606,0.0003047309,0.0001773096],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00007853685,0.00008928448,0.4871582,0.001553037,0.00003046618,0.000002135793,0.0007808494,0.1634511,0.008404022,0.000001879332,0.00007983669,0.3383706],"study_design_scores_gemma":[0.0005965048,0.0002569836,0.2793967,0.002460521,0.00006269554,5.290544e-7,0.0002558201,0.6670381,0.04828099,0.001103261,0.0002210794,0.000326874],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9140421,0.0005354052,0.08109136,0.0002976996,0.0009885384,0.002399635,0.00004694083,0.0002354769,0.0003628216],"genre_scores_gemma":[0.9764419,0.00005111522,0.02126976,0.000001151525,0.0003966927,0.0005257073,0.0001628329,0.00007787999,0.001072961],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5035869,"threshold_uncertainty_score":0.8953661,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3481346422124819,"score_gpt":0.5447860930808068,"score_spread":0.1966514508683249,"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."}}