{"id":"W2001599472","doi":"10.1007/s11219-010-9128-1","title":"An industrial case study of classifier ensembles for locating software defects","year":2011,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":67,"is_retracted":false,"has_abstract":false,"ca_institutions":"Toronto Metropolitan University","funders":"Sabancı Üniversitesi","keywords":"Computer science; Software; Classifier (UML); False alarm; Software quality; Data mining; Domain (mathematical analysis); Ensemble learning; Embedded software; Machine learning; Artificial intelligence; Reliability engineering; Software development; Engineering; Operating system","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":[0.003847868,0.0006649975,0.0006426304,0.001565289,0.001041399,0.001129865,0.001910651,0.002095669,0.00158533],"category_scores_gemma":[0.01342944,0.0003112817,0.0005537263,0.001583758,0.0006263921,0.001069658,0.0007609113,0.0008757025,0.0005023738],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007788204,"about_ca_system_score_gemma":0.000598967,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003948894,"about_ca_topic_score_gemma":0.006650547,"domain_scores_codex":[0.9974268,0.0009018117,0.0001606386,0.0003916862,0.0009275355,0.0001914634],"domain_scores_gemma":[0.9815496,0.01188258,0.0006998042,0.002517277,0.002952266,0.0003984969],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.002307813,0.003889282,0.1112214,0.0008313779,0.0003215936,0.009728532,0.002760554,0.2161428,0.0465234,0.004251115,0.007535235,0.594487],"study_design_scores_gemma":[0.0003210168,0.004064778,0.03337085,0.0001018265,0.0002979226,0.006298017,0.002249875,0.8759122,0.06423763,0.004433098,0.008612101,0.0001006601],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9270113,0.0003644177,0.0681788,0.0003270322,0.00004784302,0.0001885725,0.0001990621,0.0005432543,0.003139879],"genre_scores_gemma":[0.9472494,0.000100067,0.05086465,0.00003558389,0.00001198632,0.00003529698,0.000156882,0.00004559547,0.001500491],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003948894,"threshold_uncertainty_score":0.02034968,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2236840213549557,"score_gpt":0.3779922822970453,"score_spread":0.1543082609420896,"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."}}