"Skalsk\u00E1, Hana" . "691751" . "Znalosti 2001" . "Porovn\u00E1n\u00ED kvality klasifika\u010Dn\u00EDch algoritm\u016F v kontextu managementu znalost\u00ED"@cs . "Comparison of Classification Algorithms in the Contect of Knowledge Management"@en . "Klasifikace prvk\u016F a v\u00FDb\u011Br vhodn\u00E9ho klasifika\u010Dn\u00EDho modelu jsou \u010Dast\u00E9 typy \u00FAloh, kter\u00E9 jsou \u0159e\u0161eny v r\u00E1mci managementu znalost\u00ED. Doporu\u010Denou m\u00EDrou kvality p\u0159i klasifikaci do dvou skupin je ROC k\u0159ivka. Konvexn\u00ED obal ROCCH nad ROC k\u0159ivkami jednotliv\u00FDch klasifika\u010Dn\u00EDch algoritm\u016F umo\u017En\u00ED sestavit hybridn\u00ED klasifika\u010Dn\u00ED pravidlo, kter\u00E9 je robustn\u00ED s ohledem na zm\u011Bny pravd\u011Bpodobnost\u00ED v zastoupen\u00ED skupin a zm\u011Bny cen chybn\u00E9 klasifikace. Pro data z datov\u00E9ho depozit\u00E1\u0159e je \u0159e\u0161ena mo\u017Enost stanoven\u00ED ROC a porovn\u00E1n\u00ED model\u016F line\u00E1rn\u00ED diskrimina\u010Dn\u00ED funkce, logistick\u00E9 regrese a rozhodovac\u00EDch strom\u016F C&RT a CART. K odhadu prediktivn\u00ED validity byl pou\u017Eit nez\u00E1visl\u00FD testovac\u00ED soubor. Modely byly sestaveny na u\u010D\u00EDc\u00EDm souboru o rozsahu 30162 prvk\u016F. P\u0159esnost odhadnuta na testovac\u00EDm souboru 15060 prvk\u016F. Bylo mo\u017Eno stanovit lok\u00E1ln\u011B optim\u00E1ln\u00ED klasifika\u010Dn\u00ED pravidla (logistick\u00E1 regrese a C&RT)." . "Praha" . "Praha" . "0"^^ . . . "1"^^ . "274;279" . "0"^^ . "1"^^ . "RIV/62690094:18450/01:6418" . . "Vysok\u00E1 \u0161kola ekonomick\u00E1 v Praze" . "80-245-0190-2" . . "6"^^ . . . . "2001-01-01+01:00"^^ . . "[340ECA62385B]" . "18450" . . . . "Classification of items and selection the best classification models are very often solved in the frame of knowledge management tasks. ROC curve is reccomended as an available measure the quality of models in the two-groups classification problems. Convex hull ROCCH under ROC curves of different models was used for setting the classification rule that is robust to changes in probabilities of both groups and (or) changes in missclassification costs. ROC curves of linear discriminant function, logistic regression, C&RT and QUEST models were compared on data from data repository. Estimate of predictive validity was based on independent test sample. Sample sizes were 30162 and 15060 respectively for training and test samples."@en . "Classification Error;ROC curves;ROCCH"@en . "Porovn\u00E1n\u00ED kvality klasifika\u010Dn\u00EDch algoritm\u016F v kontextu managementu znalost\u00ED" . "Comparison of Classification Algorithms in the Contect of Knowledge Management"@en . . . "Z(MSM 184500001)" . "Porovn\u00E1n\u00ED kvality klasifika\u010Dn\u00EDch algoritm\u016F v kontextu managementu znalost\u00ED" . "Porovn\u00E1n\u00ED kvality klasifika\u010Dn\u00EDch algoritm\u016F v kontextu managementu znalost\u00ED"@cs . . . . "RIV/62690094:18450/01:6418!RIV/2002/MSM/184502/N" .