{"id":"W4388073046","doi":"10.5070/sr33162445","title":"[SoK] Evaluations in Industrial Intrusion Detection Research","year":2023,"lang":"en","type":"article","venue":"Journal of Systems Research","topic":"Network Security and Intrusion Detection","field":"Computer Science","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"Deutsche Forschungsgemeinschaft","keywords":"Benchmarking; Intrusion detection system; Field (mathematics); Data science; Computer science; Risk analysis (engineering); Computer security; Business; Marketing","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.03300521,0.001422369,0.001095916,0.01857952,0.001567337,0.005973763,0.002057874,0.001548638,0.009713993],"category_scores_gemma":[0.1456327,0.0003538732,0.001322719,0.02370755,0.001328808,0.007430866,0.002832912,0.00116823,0.005059199],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002547416,"about_ca_system_score_gemma":0.003135914,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003431839,"about_ca_topic_score_gemma":0.004739964,"domain_scores_codex":[0.9320266,0.03108559,0.009136726,0.003830876,0.02253565,0.001384509],"domain_scores_gemma":[0.8445209,0.08683398,0.01325859,0.01702839,0.03639838,0.001959743],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.002099582,0.0007833946,0.05797141,0.03362737,0.001855808,0.0004313623,0.001850797,0.01471248,0.006166217,0.04329697,0.278095,0.5591096],"study_design_scores_gemma":[0.0006819317,0.001928476,0.09931553,0.01031933,0.001740729,0.001383197,0.003773909,0.04456994,0.02114853,0.03878651,0.7760093,0.0003425481],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"other","genre_gemma":"review","genre_scores_codex":[0.2617846,0.1062986,0.07226171,0.01294114,0.003746446,0.005386425,0.2163359,0.0112794,0.3099658],"genre_scores_gemma":[0.551676,0.02597404,0.1134658,0.002543695,0.001120033,0.004216885,0.2842152,0.002413413,0.01437487],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.9669948,"threshold_uncertainty_score":0.1745503,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3379895569946412,"score_gpt":0.4654608749629295,"score_spread":0.1274713179682883,"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."}}