cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Try the Materials Informatics Toolkit, which is designed to easily handle SMILES data. This and other helpful add-ins are available in the JMP® Marketplace
%3CLINGO-SUB%20id%3D%22lingo-sub-433903%22%20slang%3D%22zh-TW%22%20mode%3D%22NONE%22%20mode%3D%22NONE%22%20mode%3D%22NONE%22%3EUtilisation%20de%20JMP%20pour%20purifier%20les%20valeurs%20aberrantes%2C%20en%20prenant%20comme%20exemple%20l'analyse%20des%20semi-conducteurs%20des%20donn%C3%A9es%20WAT%3C%2FLINGO-SUB%3E%3CLINGO-BODY%20id%3D%22lingo-body-433903%22%20slang%3D%22zh-TW%22%20mode%3D%22NONE%22%20mode%3D%22NONE%22%20mode%3D%22NONE%22%3E%3CP%3ELorsque%20vous%20analysez%20des%20donn%C3%A9es%2C%20quelle%20est%20la%20premi%C3%A8re%20%C3%A9tape%26nbsp%3B%3F%3C%2FP%3E%0A%3CP%3ESouhaitez-vous%20utiliser%20le%20puissant%20g%C3%A9n%C3%A9rateur%20de%20graphiques%20interactif%20JMP%20(Graph%20Builder)%20pour%20l'analyse%20visuelle%26nbsp%3B%3F%3C%2FP%3E%0A%3CP%3EOu%20utiliser%20la%20plate-forme%20de%20distribution%20(distribution)%20de%20JMP%20pour%20effectuer%20une%20analyse%20des%20statistiques%20narratives%26nbsp%3B%3F%3C%2FP%3E%0A%3CP%3EOu%20utiliser%20les%20divers%20outils%20de%20mod%C3%A9lisation%20pr%C3%A9dictive%20de%20JMP%20pour%20trouver%20le%20mod%C3%A8le%20pr%C3%A9dictif%20le%20plus%20appropri%C3%A9%26nbsp%3B%3F%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3EBien%20qu'il%20s'agisse%20de%20fonctions%20JMP%20bien%20connues%2C%20comme%20premi%C3%A8re%20%C3%A9tape%20de%20l'utilisation%20de%20l'analyse%20JMP%2C%20en%20plus%20des%20plates-formes%20mentionn%C3%A9es%20ci-dessus%2C%20vous%20pouvez%20envisager%20d'utiliser%20Explorer%20les%20valeurs%20aberrantes%20(explorer%20les%20valeurs%20aberrantes)%20et%20Explorer%20les%20valeurs%20manquantes%20(explorer%20les%20valeurs%20manquantes)%20Plate-forme%2C%20nettoyez%20et%20organisez%20vos%20informations.%3C%2FP%3E%0A%3CP%3EVous%20pensez%20peut-%C3%AAtre%20qu'il%20est%20compr%C3%A9hensible%20d'explorer%20les%20valeurs%20manquantes.%20Apr%C3%A8s%20tout%2C%20il%20est%20raisonnable%20de%20supprimer%20d'abord%20les%20valeurs%20nulles%20pour%20une%20analyse%20ult%C3%A9rieure.%20Mais%20pourquoi%20devez-vous%20%C3%A9galement%20exclure%20les%20valeurs%20aberrantes%26nbsp%3B%3F%20Si%20ces%20valeurs%20aberrantes%20sont%20repr%C3%A9sentatives%20de%20la%20population%20de%20donn%C3%A9es%2C%20cela%20affectera-t-il%20notre%20analyse%26nbsp%3B%3F%20De%20plus%2C%20les%20valeurs%20aberrantes%20sont%20difficiles%20%C3%A0%20capter%2C%20ne%20pouvons-nous%20pas%20d'abord%20ignorer%20cet%20effet%26nbsp%3B%3F%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3EPour%20clarifier%20ces%20questions%2C%20nous%20devons%20savoir%20ce%20qu'est%20une%20valeur%20aberrante.%3C%2FP%3E%0A%3CH2%20id%3D%22toc-hId-1704708040%22%20id%3D%22toc-hId-1704729303%22%20id%3D%22toc-hId-1704729303%22%20id%3D%22toc-hId-1704729303%22%20id%3D%22toc-hId-1704729303%22%3E%3CSTRONG%3E%3CFONT%20size%3D%224%22%3EQuelles%20sont%20les%20valeurs%20aberrantes%26nbsp%3B%3F%3C%2FFONT%3E%3C%2FSTRONG%3E%3C%2FH2%3E%0A%3CP%3EDiscutons%20d'abord%20de%20la%20situation%20unidimensionnelle%20et%20g%C3%A9n%C3%A9rons%20au%20hasard%201000%20valeurs%20de%20la%20distribution%20normale%20standard%20Nor%20(0%2C%201)%2C%20comme%20le%20montre%20la%20figure%201%2C%20o%C3%B9%20les%20points%20marqu%C3%A9s%20en%20rouge%20sont%20des%20valeurs%20aberrantes%20au-del%C3%A0%20de%20la%20plage%20de%20%5B-%204%2C4%5D%2C%20tels%20que%20la%20probabilit%C3%A9%20d'occurrence%20est%20de%200%2C0063%26nbsp%3B%25%2C%20ce%20qui%20est%20une%20tr%C3%A8s%20faible%20probabilit%C3%A9%2C%20nous%20consid%C3%A9rons%20donc%20les%20points%20qui%20se%20produisent%20dans%20une%20si%20petite%20zone%20de%20probabilit%C3%A9%20comme%20des%20valeurs%20aberrantes.%3C%2FP%3E%0A%3CP%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%20lia-image-align-inline%22%20image-alt%3D%22Michelle_Wu_0-1636341617235.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_0-1636341617235.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_0-1636341617235.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_0-1636341617235.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3Cspan%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_0-1636341617235.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3Cimg%20src%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2Fimage%2Fserverpage%2Fimage-id%2F37344iE41D1030572A641A%2Fimage-size%2Fmedium%3Fv%3Dv2%26amp%3Bpx%3D400%22%20role%3D%22button%22%20title%3D%22Michelle_Wu_0-1636341617235.png%22%20alt%3D%22Michelle_Wu_0-1636341617235.png%22%20%2F%3E%3C%2Fspan%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3EFigure%201%20Carte%20de%20points%20de%20distribution%20normale%20g%C3%A9n%C3%A9r%C3%A9e%20al%C3%A9atoirement%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3EEnsuite%2C%20nous%20examinons%20la%20situation%20de%20deux%20variables%2C%20en%20utilisant%20l'analyse%20du%20g%C3%A9n%C3%A9rateur%20de%20graphes%2C%20que%20les%20points%20de%20donn%C3%A9es%20aient%20ou%20non%20la%20m%C3%AAme%20moyenne%20et%20la%20m%C3%AAme%20variance%20entre%20les%20deux%20variables%20(X2%20vs.%20X1)%20ou%20avec%20des%20moyennes%20et%20variances%20diff%C3%A9rentes%20(X3%20vs.%20X1)%2C%20comme%20le%20montre%20la%20figure%202%20ci-dessous%2C%20nous%20pouvons%20voir%20%C3%A0%20l'%C5%93il%20nu%20que%20ces%20points%20aberrants%20(points%20rouges)%20semblent%20avoir%20une%20tendance%20diff%C3%A9rente%20des%20autres%20points%20et%20sont%20%C3%A9loign%C3%A9s%20du%20point%20du%20centre%20de%20donn%C3%A9es.%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%20lia-image-align-inline%22%20image-alt%3D%22Michelle_Wu_1-1636341617245.png%22%20style%3D%22width%3A%20360px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_1-1636341617245.png%22%20style%3D%22width%3A%20360px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_1-1636341617245.png%22%20style%3D%22width%3A%20360px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_1-1636341617245.png%22%20style%3D%22width%3A%20360px%3B%22%3E%3Cspan%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_1-1636341617245.png%22%20style%3D%22width%3A%20360px%3B%22%3E%3Cimg%20src%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2Fimage%2Fserverpage%2Fimage-id%2F37345iBB467C93CE4572D7%2Fimage-dimensions%2F360x280%3Fv%3Dv2%22%20width%3D%22360%22%20height%3D%22280%22%20role%3D%22button%22%20title%3D%22Michelle_Wu_1-1636341617245.png%22%20alt%3D%22Michelle_Wu_1-1636341617245.png%22%20%2F%3E%3C%2Fspan%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%20lia-image-align-inline%22%20image-alt%3D%22Michelle_Wu_2-1636341617252.png%22%20style%3D%22width%3A%20363px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_2-1636341617252.png%22%20style%3D%22width%3A%20363px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_2-1636341617252.png%22%20style%3D%22width%3A%20363px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_2-1636341617252.png%22%20style%3D%22width%3A%20363px%3B%22%3E%3Cspan%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_2-1636341617252.png%22%20style%3D%22width%3A%20363px%3B%22%3E%3Cimg%20src%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2Fimage%2Fserverpage%2Fimage-id%2F37346i0CFCCE4B2BC66B7B%2Fimage-dimensions%2F363x280%3Fv%3Dv2%22%20width%3D%22363%22%20height%3D%22280%22%20role%3D%22button%22%20title%3D%22Michelle_Wu_2-1636341617252.png%22%20alt%3D%22Michelle_Wu_2-1636341617252.png%22%20%2F%3E%3C%2Fspan%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FP%3E%0A%3CP%3EFigure%202%20(%C3%A0%20gauche)%202%20variables%20ont%20le%20m%C3%AAme%20nuage%20de%20points%20de%20moyenne%20et%20de%20variance%20(%C3%A0%20droite)%202%20variables%20ont%20un%20nuage%20de%20points%20de%20moyenne%20et%20de%20variance%20diff%C3%A9rents%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3EEt%20lorsqu'il%20existe%20une%20forte%20corr%C3%A9lation%20entre%20les%20deux%20variables%2C%20on%20peut%20%C3%A9galement%20constater%20que%20certaines%20valeurs%20aberrantes%20sont%20%C3%A9loign%C3%A9es%20d'autres%20points%20d'%C3%A9chantillonnage%20(points%20rouges%20sur%20la%20figure%203).%20De%20plus%2C%20en%20utilisant%20la%20fonction%20histogram%20boarder%20de%20la%20plateforme%20Fit%20Y%20by%20X%20pour%20observer%20la%20distribution%20des%20deux%20variables%20(Figure%203)%2C%20nous%20avons%20constat%C3%A9%20que%20si%20nous%20nous%20concentrons%20uniquement%20sur%20une%20seule%20variable%2C%20nous%20ne%20pourrons%20pas%20d%C3%A9tecter%20certaines%20valeurs%20aberrantes%2C%20comme%20le%20montre%20la%20figure%20ci-dessous%2C%20qui%20est%20%C3%A9galement%20un%20Un%20bon%20exemple%20montre%20que%20lorsque%20nous%20consid%C3%A9rons%20les%20valeurs%20aberrantes%20multivari%C3%A9es%2C%20nous%20capterons%20certaines%20valeurs%20aberrantes%20potentielles%20et%20%C3%A9viterons%20les%20erreurs%20de%20jugement%20en%20observant%20et%20en%20contr%C3%B4lant%20uniquement%20une%20seule%20variable.%20Heureusement%2C%20JMP%20fournit%20une%20telle%20plate-forme%20d'observation%20des%20valeurs%20aberrantes%20multivari%C3%A9es.%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%20lia-image-align-inline%22%20image-alt%3D%22Michelle_Wu_3-1636341617255.png%22%20style%3D%22width%3A%20450px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_3-1636341617255.png%22%20style%3D%22width%3A%20450px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_3-1636341617255.png%22%20style%3D%22width%3A%20450px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_3-1636341617255.png%22%20style%3D%22width%3A%20450px%3B%22%3E%3Cspan%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_3-1636341617255.png%22%20style%3D%22width%3A%20450px%3B%22%3E%3Cimg%20src%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2Fimage%2Fserverpage%2Fimage-id%2F37347i3E2204F7D17B9937%2Fimage-dimensions%2F450x360%3Fv%3Dv2%22%20width%3D%22450%22%20height%3D%22360%22%20role%3D%22button%22%20title%3D%22Michelle_Wu_3-1636341617255.png%22%20alt%3D%22Michelle_Wu_3-1636341617255.png%22%20%2F%3E%3C%2Fspan%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FP%3E%0A%3CP%3EFigure%203%20Nuage%20de%20points%20avec%20forte%20corr%C3%A9lation%20entre%20deux%20variables%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3E%C3%80%20ce%20stade%2C%20nous%20pouvons%20comprendre%20le%20concept%20de%20valeurs%20aberrantes%2C%20qui%20consiste%20%C3%A0%20trouver%20des%20points%20plus%20%C3%A9loign%C3%A9s%20du%20centre%20de%20l'%C3%A9chantillon%20que%20les%20autres%20points%20de%20l'%C3%A9chantillon%2C%20ou%20lorsque%20l'emplacement%20du%20point%20ne%20correspond%20pas%20%C3%A0%20la%20%22tendance%22%20globale%20de%20l'%C3%A9chantillon.%20Fait%20int%C3%A9ressant%2C%20les%20valeurs%20aberrantes%20ne%20repr%C3%A9sentent%20pas%20des%20points%20probl%C3%A9matiques%2C%20mais%20repr%C3%A9sentent%20un%20%C3%A9cart%20par%20rapport%20aux%20autres%20%C3%A9chantillons%20sans%20la%20m%C3%AAme%20%22tendance%22.%20Par%20cons%C3%A9quent%2C%20les%20valeurs%20aberrantes%20peuvent%20repr%C3%A9senter%20des%20personnages%20probl%C3%A9matiques%20dans%20une%20bonne%20soci%C3%A9t%C3%A9%2C%20ou%20elles%20peuvent%20repr%C3%A9senter%20un%20flux%20propre%20dans%20une%20ville%20m%C3%A9chante%2C%20selon%20la%20provenance%20de%20vos%20%C3%A9chantillons.%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CH2%20id%3D%22toc-hId--847448921%22%20id%3D%22toc-hId--847427658%22%20id%3D%22toc-hId--847427658%22%20id%3D%22toc-hId--847427658%22%20id%3D%22toc-hId--847427658%22%3E%3CSTRONG%3E%3CFONT%20size%3D%224%22%3EFaire%20bon%20usage%20de%20JMP%20pour%20analyser%20les%20donn%C3%A9es%20WAT%20et%20r%C3%A9soudre%20les%20probl%C3%A8mes%20de%20traitement%20des%20semi-conducteurs%3C%2FFONT%3E%3C%2FSTRONG%3E%3C%2FH2%3E%0A%3CP%3ECela%20dit%2C%20nous%20utilisons%20les%20donn%C3%A9es%20d'un%20test%20%C3%A9lectrique%20de%20fabrication%20de%20semi-conducteurs%20%C3%A0%20titre%20d'illustration.%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3ELes%20param%C3%A8tres%20des%20composants%20de%20la%20fabrication%20de%20plaquettes%20semi-conductrices%20ont%20leurs%20caract%C3%A9ristiques%20%C3%A9lectroniques%20correspondantes%2C%20de%20sorte%20que%20la%20qualit%C3%A9%20du%20processus%20correspondant%20peut%20%C3%AAtre%20d%C3%A9duite%20%C3%A0%20travers%20les%20r%C3%A9sultats%20des%20tests%20%C3%A9lectriques%20de%20ces%20param%C3%A8tres.%20Les%20fonderies%20de%20plaquettes%20appellent%20g%C3%A9n%C3%A9ralement%20le%20test%20des%20param%C3%A8tres%20%C3%A9lectriques%20WAT%20(Wafer%20Acceptance%20Test)%2C%20et%20leurs%20clients%20doivent%20v%C3%A9rifier%20les%20donn%C3%A9es%20du%20test%20WAT%20pour%20confirmer%20la%20qualit%C3%A9%20de%20fabrication.%20Dans%20l'%C3%A9tape%20de%20R%26amp%3BD%20du%20nouveau%20proc%C3%A9d%C3%A9%2C%20il%20est%20n%C3%A9cessaire%20de%20collecter%20une%20grande%20quantit%C3%A9%20de%20donn%C3%A9es%20WAT%20pour%20r%C3%A9soudre%20les%20probl%C3%A8mes%20de%20proc%C3%A9d%C3%A9%20ou%20trouver%20la%20configuration%20optimale%20des%20facteurs.%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3ENous%20utilisons%20un%20ensemble%20de%20donn%C3%A9es%20WAT%20simples%20comme%20d%C3%A9monstration%2C%20qui%20contient%20des%20donn%C3%A9es%20de%20param%C3%A8tres%20d'observation%2010.%20Lorsque%20nous%20analysons%20directement%20les%20donn%C3%A9es%20de%20la%20ligne%20de%20production%203%2C%20nous%20pouvons%20utiliser%20la%20fonction%20Analyser%26gt%3B%20Distribution%20pour%20cr%C3%A9er%20l'histogramme%20correspondant%20et%20la%20bo%C3%AEte%20%C3%A0%20moustaches%20pour%20voir%20si%20s'il%20y%20a%20un%20%C3%A9cart.%20Valeurs%20de%20groupe%2C%20comme%20le%20montre%20la%20figure%204%20(%C3%A0%20gauche)%2C%20la%20ligne%20rouge%20sur%20la%20bo%C3%AEte%20des%20valeurs%20aberrantes%20marque%20la%20zone%20la%20plus%20concentr%C3%A9e%20couvrant%2050%26nbsp%3B%25%2C%20%C3%A0%20partir%20de%20laquelle%20nous%20pouvons%20trouver%20que%20les%20donn%C3%A9es%20sont%20concentr%C3%A9es%20dans%20la%20section%20m%C3%A9diane.%20De%20m%C3%AAme%2C%20vous%20pouvez%20%C3%A9galement%20utiliser%20la%20fonction%20d'histogramme%20de%20Graph%20Builder%20pour%20terminer%20le%20dessin%20de%20l'histogramme%2C%20comme%20illustr%C3%A9%20%C3%A0%20la%20figure%204%20(%C3%A0%20droite).%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%20lia-image-align-inline%22%20image-alt%3D%22Michelle_Wu_4-1636341617257.png%22%20style%3D%22width%3A%20251px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_4-1636341617257.png%22%20style%3D%22width%3A%20251px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_4-1636341617257.png%22%20style%3D%22width%3A%20251px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_4-1636341617257.png%22%20style%3D%22width%3A%20251px%3B%22%3E%3Cspan%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_4-1636341617257.png%22%20style%3D%22width%3A%20251px%3B%22%3E%3Cimg%20src%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2Fimage%2Fserverpage%2Fimage-id%2F37350i704A49E62E87FBEE%2Fimage-dimensions%2F251x530%3Fv%3Dv2%22%20width%3D%22251%22%20height%3D%22530%22%20role%3D%22button%22%20title%3D%22Michelle_Wu_4-1636341617257.png%22%20alt%3D%22Michelle_Wu_4-1636341617257.png%22%20%2F%3E%3C%2Fspan%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%26nbsp%3B%20%26nbsp%3B%20%26nbsp%3B%26nbsp%3B%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%20lia-image-align-inline%22%20image-alt%3D%22Michelle_Wu_5-1636341617260.png%22%20style%3D%22width%3A%20361px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_5-1636341617260.png%22%20style%3D%22width%3A%20361px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_5-1636341617260.png%22%20style%3D%22width%3A%20361px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_5-1636341617260.png%22%20style%3D%22width%3A%20361px%3B%22%3E%3Cspan%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_5-1636341617260.png%22%20style%3D%22width%3A%20361px%3B%22%3E%3Cimg%20src%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2Fimage%2Fserverpage%2Fimage-id%2F37349i0731F301D779A7AA%2Fimage-dimensions%2F361x300%3Fv%3Dv2%22%20width%3D%22361%22%20height%3D%22300%22%20role%3D%22button%22%20title%3D%22Michelle_Wu_5-1636341617260.png%22%20alt%3D%22Michelle_Wu_5-1636341617260.png%22%20%2F%3E%3C%2Fspan%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FP%3E%0A%3CP%3EFigure%204%20(%C3%A0%20gauche)%20Histogramme%20de%20la%20plate-forme%20Distribution%26nbsp%3B%3B%20bo%C3%AEte%20%C3%A0%20moustaches%20des%20valeurs%20aberrantes%20(%C3%A0%20droite)%20Bo%C3%AEte%20%C3%A0%20moustaches%20du%20g%C3%A9n%C3%A9rateur%20de%20graphes%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3ESelon%20le%20test%20de%20valeur%20aberrante%20de%20la%20plage%20quantile%2C%20nous%20avons%20constat%C3%A9%20qu'il%20existe%20une%20valeur%20aberrante%20(point%20rouge)%20dans%20le%20param%C3%A8tre%201.%20Utilisez%20la%20fonction%20Valeurs%20aberrantes%20de%20la%20plage%20quantile%20dans%20la%20plate-forme%20Analyze%26gt%3BScreening%26gt%3BExplore%20Outliers%20de%20JMP%20pour%20trouver%20sa%20valeur%20aberrante%20(Figure%205)%2C%20et%20des%20actions%20telles%20que%20car%20le%20marquage%2C%20l'exclusion%20et%20le%20traitement%20des%20donn%C3%A9es%20comme%20des%20valeurs%20manquantes%20peuvent%20%C3%AAtre%20effectu%C3%A9s.%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3EDans%20le%20m%C3%AAme%20temps%2C%20vous%20pouvez%20%C3%A9galement%20ajuster%20les%20r%C3%A8gles%20de%20jugement%20des%20valeurs%20aberrantes.%20Vous%20pouvez%20ajuster%20la%20valeur%20du%20centile%20de%20queue%20(Quantile%20de%20queue)%20et%20son%20multiple%20(Q)%2C%20et%20utiliser%20le%20test%20de%20valeur%20aberrante%20IQR%20commun.%20La%20limite%20sup%C3%A9rieure%20est%20Q3%20%2B%201%2C5*.%20IQR%2C%20la%20limite%20sup%C3%A9rieure%20Prenant%20Q1-1.5*IQR%20comme%20exemple%2C%20Tail%20Quantile%20est%200.25%2C%20et%20Q%20est%201.5.%20Et%20lorsque%20vos%20donn%C3%A9es%20ne%20sont%20pas%20conformes%20%C3%A0%20la%20distribution%20normale%2C%20vous%20pouvez%20envisager%20d'utiliser%20Robust%20Fit%20Outliers%20comme%20plate-forme%20de%20d%C3%A9pistage%20des%20valeurs%20aberrantes.%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%20lia-image-align-inline%22%20image-alt%3D%22Michelle_Wu_6-1636341617263.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_6-1636341617263.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_6-1636341617263.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_6-1636341617263.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3Cspan%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_6-1636341617263.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3Cimg%20src%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2Fimage%2Fserverpage%2Fimage-id%2F37352i604BC27C146CAE3C%2Fimage-size%2Fmedium%3Fv%3Dv2%26amp%3Bpx%3D400%22%20role%3D%22button%22%20title%3D%22Michelle_Wu_6-1636341617263.png%22%20alt%3D%22Michelle_Wu_6-1636341617263.png%22%20%2F%3E%3C%2Fspan%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FP%3E%0A%3CP%3EFigure%205%20Quantile%20Range%20Outliers%20dans%20la%20plate-forme%20Explore%20Outliers%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3EMais%20lorsque%20nous%20l'examinons%20plus%20en%20d%C3%A9tail%20et%20incluons%20les%20donn%C3%A9es%20de%20toutes%20les%20lignes%20de%20production%2C%20vous%20constaterez%20que%20la%20ligne%20de%20production%203%20a%20un%20r%C3%A9sultat%20de%20test%20de%20param%C3%A8tre%201%20plus%20%C3%A9lev%C3%A9%2C%20comme%20le%20montre%20la%20figure%206.%20La%20valeur%20aberrante%20d'origine%20de%208%2C01%20sur%20la%20ligne%20de%20production%203%20est%20plus%20en%20ligne%20avec%20la%20plupart%20des%20donn%C3%A9es%20Cependant%2C%20les%20valeurs%20aberrantes%20r%C3%A9elles%20sont%20converties%20aux%20trois%20valeurs%20les%20plus%20%C3%A9lev%C3%A9es%20de%20la%20ligne%20de%20production%203%20(points%20rouges%20sur%20la%20figure%206).%20Ce%20r%C3%A9sultat%20fait%20%C3%A9galement%20%C3%A9cho%20%C3%A0%20l'importance%20de%20la%20source%20des%20%C3%A9chantillons%20susmentionn%C3%A9s.%20repr%C3%A9senter%20la%20population%20maternelle%20est%20essentiel%20pour%20un%20d%C3%A9pistage%20correct%20Les%20valeurs%20aberrantes%20jouent%20un%20r%C3%B4le%20central.%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%20lia-image-align-inline%22%20image-alt%3D%22Michelle_Wu_7-1636341617265.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_7-1636341617265.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_7-1636341617265.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_7-1636341617265.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3Cspan%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_7-1636341617265.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3Cimg%20src%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2Fimage%2Fserverpage%2Fimage-id%2F37351i3B6C9ED8B7BE557D%2Fimage-size%2Fmedium%3Fv%3Dv2%26amp%3Bpx%3D400%22%20role%3D%22button%22%20title%3D%22Michelle_Wu_7-1636341617265.png%22%20alt%3D%22Michelle_Wu_7-1636341617265.png%22%20%2F%3E%3C%2Fspan%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FP%3E%0A%3CP%3EFigure%206%20Bo%C3%AEte%20%C3%A0%20moustaches%20de%20toutes%20les%20lignes%20de%20production%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3ELors%20de%20la%20prise%20en%20compte%20de%20l'influence%20de%20tous%20les%20param%C3%A8tres%20et%20de%20la%20recherche%20de%20valeurs%20aberrantes%20sous%20l'influence%20de%20tous%20les%20param%C3%A8tres%2C%20JMP%20fournit%20%C3%A9galement%20des%20fonctions%20Robust%20PCA%20Outliers%20ou%20K%20Nearest%20Neighbor%20Outliers%20que%20vous%20pouvez%20utiliser.%20Vous%20pouvez%20trouver%20ces%20deux%20plates-formes%20fonctionnelles%20sous%20Analyser%26gt%3B%20D%C3%A9pistage%26gt%3B%20Explorer%20les%20valeurs%20aberrantes%20Ici%2C%20utilisez%20la%20plate-forme%20K%20Nearest%20Neighbor%20Outliers%20pour%20trouver%20des%20valeurs%20aberrantes%20et%20utilisez%20le%20composite%20pour%20s%C3%A9lectionner%20des%20points%20d'%C3%A9chantillonnage%20avec%20des%20distances%20plus%20%C3%A9lev%C3%A9es%20et%20les%20marquer%20comme%20des%20valeurs%20aberrantes%20(Figure%207).%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%20lia-image-align-inline%22%20image-alt%3D%22Michelle_Wu_8-1636341617275.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_8-1636341617275.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_8-1636341617275.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_8-1636341617275.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3Cspan%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_8-1636341617275.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3Cimg%20src%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2Fimage%2Fserverpage%2Fimage-id%2F37353i3CA3D3531BCC66EF%2Fimage-size%2Fmedium%3Fv%3Dv2%26amp%3Bpx%3D400%22%20role%3D%22button%22%20title%3D%22Michelle_Wu_8-1636341617275.png%22%20alt%3D%22Michelle_Wu_8-1636341617275.png%22%20%2F%3E%3C%2Fspan%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FP%3E%0A%3CP%3EFigure%207%20Carte%20de%20distance%20K%20Nearest%20Neighbor%20Outliers%20sous%20la%20plate-forme%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3EEn%20confirmant%20la%20position%20de%20distribution%20de%20ces%20points%20de%20donn%C3%A9es%20s%C3%A9lectionn%C3%A9s%20comme%20valeurs%20aberrantes%20sur%20chaque%20param%C3%A8tre%20par%20le%20g%C3%A9n%C3%A9rateur%20de%20graphiques%2C%20on%20peut%20constater%20que%20tous%20les%20points%20n'appartiennent%20pas%20aux%20valeurs%20aberrantes%20d'un%20seul%20param%C3%A8tre%20(tels%20que%20les%20points%20de%20donn%C3%A9es%20290%2C%20295%2C%20296%2C%20297%2C%20298%20%2C%20299%20)%2C%20ce%20r%C3%A9sultat%20est%20%C3%A9galement%20coh%C3%A9rent%20avec%20la%20plate-forme%20de%20d%C3%A9pistage%20des%20valeurs%20aberrantes%20multivari%C3%A9es%20de%20JMP%20mentionn%C3%A9e%20ci-dessus%2C%20qui%20peut%20capturer%20de%20nombreux%20points%20probl%C3%A9matiques%20potentiels%20qui%20sont%20ignor%C3%A9s%20dans%20l'observation%20de%20variables%20uniques%20(Figure%208).%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%20lia-image-align-inline%22%20image-alt%3D%22Michelle_Wu_9-1636341617283.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_9-1636341617283.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_9-1636341617283.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_9-1636341617283.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3Cspan%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_9-1636341617283.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3Cimg%20src%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2Fimage%2Fserverpage%2Fimage-id%2F37356iA0F6A021CB5F9DC0%2Fimage-size%2Fmedium%3Fv%3Dv2%26amp%3Bpx%3D400%22%20role%3D%22button%22%20title%3D%22Michelle_Wu_9-1636341617283.png%22%20alt%3D%22Michelle_Wu_9-1636341617283.png%22%20%2F%3E%3C%2Fspan%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FP%3E%0A%3CP%3ELa%20figure%208%20utilise%20Boxplot%20pour%20comparer%20la%20distribution%20des%20valeurs%20aberrantes%20sur%20les%20variables%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CH2%20id%3D%22toc-hId-895361414%22%20id%3D%22toc-hId-895382677%22%20id%3D%22toc-hId-895382677%22%20id%3D%22toc-hId-895382677%22%20id%3D%22toc-hId-895382677%22%3E%3CSTRONG%3E%3CFONT%20size%3D%224%22%3ESupprimez%20les%20valeurs%20aberrantes%20et%20cr%C3%A9ez%20des%20mod%C3%A8les%20pr%C3%A9dictifs%20plus%20pr%C3%A9cis%3C%2FFONT%3E%3C%2FSTRONG%3E%3C%2FH2%3E%0A%3CP%3ENous%20savons%20que%20JMP%20dispose%20d'une%20bonne%20plate-forme%20qui%20peut%20nous%20aider%20%C3%A0%20filtrer%20les%20valeurs%20aberrantes%2C%20mais%20quel%20sera%20l'impact%20de%20ne%20pas%20filtrer%20les%20valeurs%20aberrantes%26nbsp%3B%3F%20Pour%20simplifier%2C%20regardons%20d'abord%20l'impact%20des%20valeurs%20aberrantes%20entre%20les%20deux%20param%C3%A8tres.%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3ENous%20observons%20la%20valeur%20R-carr%C3%A9%20estim%C3%A9e%20par%20la%20formule%20de%20r%C3%A9gression%20de%20l'effet%20secondaire%20entre%20le%20param%C3%A8tre%20de%20premier%20niveau%20et%20le%20deuxi%C3%A8me%20param%C3%A8tre%20des%20donn%C3%A9es%20WAT.%20Avant%20de%20supprimer%20les%20valeurs%20aberrantes%20filtr%C3%A9es%2C%20la%20valeur%20R-carr%C3%A9%20est%20de%200%2C602%2C%20comme%20le%20montre%20la%20figure%209%20(%20a)%2C%20et%20apr%C3%A8s%20suppression%20des%20valeurs%20aberrantes%2C%20la%20valeur%20R-carr%C3%A9%20est%20pass%C3%A9e%20%C3%A0%200%2C740%2C%20comme%20le%20montre%20la%20figure%209(b)%2C%20ce%20qui%20montre%20que%20la%20suppression%20des%20valeurs%20aberrantes%20peut%20construire%20un%20mod%C3%A8le%20plus%20pr%C3%A9cis%20avec%20un%20%C3%A9cart%20plus%20petit%20dans%20les%20valeurs%20estim%C3%A9es.%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3ELorsque%20nous%20utilisons%20ces%20param%C3%A8tres%20pour%20construire%20des%20r%C3%A9sultats%20de%20pr%C3%A9vision%2C%20tels%20que%20des%20pr%C3%A9visions%20de%20taux%20de%20rendement%20ou%20des%20pr%C3%A9visions%20de%20test%20CP%2FFT%2C%20nous%20pouvons%20d%C3%A9tecter%20les%20d%C3%A9fauts%20plus%20efficacement%2C%20r%C3%A9duire%20les%20co%C3%BBts%20et%20augmenter%20les%20avantages.%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%20lia-image-align-inline%22%20image-alt%3D%22Michelle_Wu_10-1636341617287.png%22%20style%3D%22width%3A%20363px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_10-1636341617287.png%22%20style%3D%22width%3A%20363px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_10-1636341617287.png%22%20style%3D%22width%3A%20363px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_10-1636341617287.png%22%20style%3D%22width%3A%20363px%3B%22%3E%3Cspan%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_10-1636341617287.png%22%20style%3D%22width%3A%20363px%3B%22%3E%3Cimg%20src%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2Fimage%2Fserverpage%2Fimage-id%2F37354iD898E2A1485E4D7D%2Fimage-dimensions%2F363x300%3Fv%3Dv2%22%20width%3D%22363%22%20height%3D%22300%22%20role%3D%22button%22%20title%3D%22Michelle_Wu_10-1636341617287.png%22%20alt%3D%22Michelle_Wu_10-1636341617287.png%22%20%2F%3E%3C%2Fspan%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%20lia-image-align-inline%22%20image-alt%3D%22Michelle_Wu_11-1636341617291.png%22%20style%3D%22width%3A%20367px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_11-1636341617291.png%22%20style%3D%22width%3A%20367px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_11-1636341617291.png%22%20style%3D%22width%3A%20367px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_11-1636341617291.png%22%20style%3D%22width%3A%20367px%3B%22%3E%3Cspan%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_11-1636341617291.png%22%20style%3D%22width%3A%20367px%3B%22%3E%3Cimg%20src%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2Fimage%2Fserverpage%2Fimage-id%2F37355iDE22203ECAC1E051%2Fimage-dimensions%2F367x301%3Fv%3Dv2%22%20width%3D%22367%22%20height%3D%22301%22%20role%3D%22button%22%20title%3D%22Michelle_Wu_11-1636341617291.png%22%20alt%3D%22Michelle_Wu_11-1636341617291.png%22%20%2F%3E%3C%2Fspan%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FP%3E%0A%3CP%3EFigure%209%20(a)%20Estimation%20par%20r%C3%A9gression%20quadratique%20sans%20valeurs%20aberrantes%20supprim%C3%A9es%20(b)%20Estimation%20par%20r%C3%A9gression%20quadratique%20avec%20suppression%20des%20valeurs%20aberrantes%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3EEffectuez%20ensuite%20un%20zoom%20avant%20et%20examinez%20tous%20les%20param%C3%A8tres%2C%20puis%20utilisez%20ces%20param%C3%A8tres%20pour%20pr%C3%A9dire%20les%20r%C3%A9sultats%20de%20la%20valeur%20Bin%20suivants%20afin%20de%20trouver%20les%20facteurs%20cl%C3%A9s.%20Et%20trouvez%20la%20meilleure%20valeur%20de%20param%C3%A8tre%20de%20facteur%20sous%20le%20param%C3%A8tre%20de%20minimisation%20de%20la%20valeur%20Bin.%20De%20la%20m%C3%AAme%20mani%C3%A8re%2C%20nous%20envisageons%20d'abord%20de%20ne%20pas%20supprimer%20les%20valeurs%20aberrantes%20pour%20trouver%20le%20mod%C3%A8le%20de%20r%C3%A9gression%20le%20plus%20appropri%C3%A9.En%20utilisant%20la%20plate-forme%20de%20mod%C3%A8le%20Fit%2C%20nous%20obtenons%20un%20mod%C3%A8le%20de%20r%C3%A9gression%20avec%20un%20R-carr%C3%A9%20d'environ%200%2C971%2C%20comprenant%20de%20nombreux%20%C3%A9l%C3%A9ments%20d'effet%20de%20facteur%2C%20parmi%20lesquels%20nous%20pouvons%20voir%20beaucoup%20L'effet%20du%20terme%20d'interaction%20est%20plus%20fort%20que%20l'effet%20principal%2C%20qui%20semble%20%C3%AAtre%20plus%20chaotique%2C%20et%20en%20observant%20le%20R%C3%A9siduel%20par%20Row%20Plot%2C%20on%20peut%20constater%20que%20l'%C3%A9cart%20entre%20les%20%C3%A9chantillons%20s%C3%A9lectionn%C3%A9s%20comme%20valeurs%20aberrantes%20est%20relativement%20important%20(Figure%2010).%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%20lia-image-align-inline%22%20image-alt%3D%22Michelle_Wu_12-1636341617296.png%22%20style%3D%22width%3A%20383px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_12-1636341617296.png%22%20style%3D%22width%3A%20383px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_12-1636341617296.png%22%20style%3D%22width%3A%20383px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_12-1636341617296.png%22%20style%3D%22width%3A%20383px%3B%22%3E%3Cspan%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_12-1636341617296.png%22%20style%3D%22width%3A%20383px%3B%22%3E%3Cimg%20src%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2Fimage%2Fserverpage%2Fimage-id%2F37357iF389718627EB7796%2Fimage-dimensions%2F383x238%3Fv%3Dv2%22%20width%3D%22383%22%20height%3D%22238%22%20role%3D%22button%22%20title%3D%22Michelle_Wu_12-1636341617296.png%22%20alt%3D%22Michelle_Wu_12-1636341617296.png%22%20%2F%3E%3C%2Fspan%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%20lia-image-align-inline%22%20image-alt%3D%22Michelle_Wu_13-1636341617307.png%22%20style%3D%22width%3A%20340px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_13-1636341617307.png%22%20style%3D%22width%3A%20340px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_13-1636341617307.png%22%20style%3D%22width%3A%20340px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_13-1636341617307.png%22%20style%3D%22width%3A%20340px%3B%22%3E%3Cspan%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_13-1636341617307.png%22%20style%3D%22width%3A%20340px%3B%22%3E%3Cimg%20src%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2Fimage%2Fserverpage%2Fimage-id%2F37359i4EBE69D7F2DA5CB6%2Fimage-dimensions%2F340x363%3Fv%3Dv2%22%20width%3D%22340%22%20height%3D%22363%22%20role%3D%22button%22%20title%3D%22Michelle_Wu_13-1636341617307.png%22%20alt%3D%22Michelle_Wu_13-1636341617307.png%22%20%2F%3E%3C%2Fspan%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FP%3E%0A%3CP%3EFigure%2010%20R%C3%A9gression%20R-carr%C3%A9%2C%20R%C3%A9siduel%20par%20diagramme%20en%20ligne%20et%20%C3%A9l%C3%A9ments%20d'effet%20de%20facteur%20affect%C3%A9s%20sans%20supprimer%20les%20valeurs%20aberrantes%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3EL'%C3%A9tape%20suivante%20consiste%20%C3%A0%20supprimer%20les%20valeurs%20aberrantes%20et%20%C3%A0%20refaire%20une%20estimation%20de%20r%C3%A9gression.%20Nous%20avons%20constat%C3%A9%20que%20non%20seulement%20le%20R-carr%C3%A9%20a%20augment%C3%A9%20%C3%A0%200%2C999%2C%20mais%20que%20les%20facteurs%20cl%C3%A9s%20ont%20%C3%A9galement%20%C3%A9t%C3%A9%20r%C3%A9duits%20%C3%A0%20quatre%20%C3%A9l%C3%A9ments%20d'effet%20de%20facteur.%20une%20diff%C3%A9rence%20significative%20a%20%C3%A9t%C3%A9%20trouv%C3%A9e%20dans%20l'%C3%A9chantillon%20(Figure%2011).%20Les%20r%C3%A9sultats%20pr%C3%A9dits%20selon%20cette%20formule%20de%20r%C3%A9gression%20seront%20%C3%A9galement%20plus%20pr%C3%A9cis%2C%20%C3%A9vitant%20les%20mauvais%20jugements.%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%20lia-image-align-inline%22%20image-alt%3D%22Michelle_Wu_14-1636341617309.png%22%20style%3D%22width%3A%20358px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_14-1636341617309.png%22%20style%3D%22width%3A%20358px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_14-1636341617309.png%22%20style%3D%22width%3A%20358px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_14-1636341617309.png%22%20style%3D%22width%3A%20358px%3B%22%3E%3Cspan%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_14-1636341617309.png%22%20style%3D%22width%3A%20358px%3B%22%3E%3Cimg%20src%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2Fimage%2Fserverpage%2Fimage-id%2F37358i412F454D0225765F%2Fimage-dimensions%2F358x216%3Fv%3Dv2%22%20width%3D%22358%22%20height%3D%22216%22%20role%3D%22button%22%20title%3D%22Michelle_Wu_14-1636341617309.png%22%20alt%3D%22Michelle_Wu_14-1636341617309.png%22%20%2F%3E%3C%2Fspan%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FP%3E%0A%3CP%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%20lia-image-align-inline%22%20image-alt%3D%22Michelle_Wu_15-1636341617310.png%22%20style%3D%22width%3A%20360px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_15-1636341617310.png%22%20style%3D%22width%3A%20360px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_15-1636341617310.png%22%20style%3D%22width%3A%20360px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_15-1636341617310.png%22%20style%3D%22width%3A%20360px%3B%22%3E%3Cspan%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_15-1636341617310.png%22%20style%3D%22width%3A%20360px%3B%22%3E%3Cimg%20src%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2Fimage%2Fserverpage%2Fimage-id%2F37360i097E45D5B4A6594A%2Fimage-dimensions%2F360x80%3Fv%3Dv2%22%20width%3D%22360%22%20height%3D%2280%22%20role%3D%22button%22%20title%3D%22Michelle_Wu_15-1636341617310.png%22%20alt%3D%22Michelle_Wu_15-1636341617310.png%22%20%2F%3E%3C%2Fspan%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FP%3E%0A%3CP%3EFigure%2011.%20La%20formule%20de%20r%C3%A9gression%20des%20valeurs%20aberrantes%20R-carr%C3%A9%2C%20R%C3%A9siduel%20par%20graphique%20en%20ligne%20et%20les%20%C3%A9l%C3%A9ments%20d'effet%20de%20facteur%20concern%C3%A9s%20ne%20sont%20pas%20supprim%C3%A9s%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3ECollectez%20un%20peu%20plus%20de%20donn%C3%A9es%20pour%20effectuer%20des%20tests%20de%20suivi%20de%20l'exactitude%20de%20la%20formule%20de%20r%C3%A9gression.%20Nous%20pouvons%20comparer%20la%20valeur%20estim%C3%A9e%20de%20la%20formule%20de%20r%C3%A9gression%20avec%20la%20valeur%20r%C3%A9elle%20de%20Bin%20sous%20la%20plate-forme%20Analyze%26gt%3BFit%20Y%20by%20X.%20Les%20nouvelles%20donn%C3%A9es%20peuvent%20toujours%20avoir%20un%20tr%C3%A8s%20bon%20niveau.%20R-carr%C3%A9%20a%20une%20pr%C3%A9cision%20de%200%2C999%20et%20aucun%20point%20particulier%20n'a%20une%20variance%20excessive.%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%20lia-image-align-inline%22%20image-alt%3D%22Michelle_Wu_16-1636341617313.png%22%20style%3D%22width%3A%20400px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_16-1636341617313.png%22%20style%3D%22width%3A%20324px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_16-1636341617313.png%22%20style%3D%22width%3A%20324px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_16-1636341617313.png%22%20style%3D%22width%3A%20324px%3B%22%3E%3Cspan%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%22Michelle_Wu_16-1636341617313.png%22%20style%3D%22width%3A%20324px%3B%22%3E%3Cimg%20src%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2Fimage%2Fserverpage%2Fimage-id%2F37361i983691B4D6D1AF11%2Fimage-size%2Fmedium%3Fv%3Dv2%26amp%3Bpx%3D400%22%20role%3D%22button%22%20title%3D%22Michelle_Wu_16-1636341617313.png%22%20alt%3D%22Michelle_Wu_16-1636341617313.png%22%20%2F%3E%3C%2Fspan%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FP%3E%0A%3CP%3EFigure%2012%20Comparaison%20de%20la%20valeur%20estim%C3%A9e%20de%20la%20formule%20de%20r%C3%A9gression%20et%20de%20la%20valeur%20Bin%20r%C3%A9elle%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3EAvec%20la%20plate-forme%20Explore%20Outliers%20de%20JMP%2C%20nous%20pouvons%20facilement%20et%20rapidement%20%C3%A9liminer%20les%20valeurs%20aberrantes%2C%20en%20%C3%A9vitant%20l'influence%20des%20valeurs%20aberrantes%20pour%20mal%20%C3%A9valuer%20la%20corr%C3%A9lation%20entre%20les%20facteurs%20ou%20faire%20de%20mauvaises%20pr%C3%A9dictions%20de%20mod%C3%A8le%2C%20conduisant%20%C3%A0%20de%20mauvaises%20estimations%20et%20d%C3%A9cisions.%20En%20tant%20qu'avant-poste%20d'analyse%2C%3CSTRONG%3E%20La%20plate-forme%20Explore%20Outliers%20de%20JMP%20peut%20traiter%20diff%C3%A9rents%20sc%C3%A9narios%2C%20qu'ils%20soient%20univari%C3%A9s%20ou%20multivari%C3%A9s%2C%20que%20les%20donn%C3%A9es%20soient%20normales%2C%20et%20peut%20proposer%20une%20plate-forme%20appropri%C3%A9e%20pour%20aider%20les%20utilisateurs%20%C3%A0%20trouver%20des%20valeurs%20aberrantes%20et%20%C3%A0%20atteindre%20rapidement%20l'objectif%20de%20nettoyage%20des%20donn%C3%A9es.%3C%2FSTRONG%3E%20.%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3ELecture%20%C3%A9tendue%20%3A%3C%2FP%3E%0A%3CP%3E%3CA%20href%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2FJMPer-Cable%2FHow-to-make-a-wafer-map-in-JMP-in-under-30-seconds%2Fba-p%2F348607%22%20target%3D%22_blank%22%3EComment%20cr%C3%A9er%20une%20carte%20de%20plaquettes%20dans%20JMP%20en%2030%20secondes%20(anglais)%3C%2FA%3E%3C%2FP%3E%0A%3CP%3E%3CA%20href%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2FJMPer-Cable%2FImage-and-wafer-analysis-using-functional-data-analysis%2Fba-p%2F225444%22%20target%3D%22_blank%22%3EAnalyse%20d'images%20et%20de%20tranches%20%C3%A0%20l'aide%20de%20l'analyse%20de%20donn%C3%A9es%20fonctionnelles%3C%2FA%3E%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CH2%20id%3D%22toc-hId--1656795547%22%20id%3D%22toc-hId--1656774284%22%20id%3D%22toc-hId--1656774284%22%20id%3D%22toc-hId--1656774284%22%20id%3D%22toc-hId--1656774284%22%3E%3CSTRONG%3E%3CFONT%20size%3D%224%22%3EInscrivez-vous%20au%2011%2F26%20Semiconductor%20Symposium%3C%2FFONT%3E%3C%2FSTRONG%3E%3C%2FH2%3E%0A%3CP%3E%26gt%3B%26gt%3B%3CSPAN%3E%20%C3%A0%20travers%3C%2FSPAN%3E%20Que%20diriez-vous%20de%20JMP%3CSPAN%3E%20Permettre%20aux%20utilisateurs%20d'obtenir%20rapidement%20et%20efficacement%20des%20combinaisons%20de%20param%C3%A8tres%20optimis%C3%A9es%2C%20am%C3%A9liorant%20ainsi%20le%20rendement%20de%20la%20ligne%20de%20production%20et%20augmentant%20la%20production%26nbsp%3B%3F%20Inscrivez-vous%20au%2011%2F26%20Semiconductor%20Symposium%2C%3C%2FSPAN%3E%20apprendre%20%C3%A0%20%C3%AAtre%20gentil%3CSPAN%3E%20Utilisez%20JMP%20pour%20l'analyse%20WAT%20afin%20d'am%C3%A9liorer%20le%20rendement%20du%20processus%20et%20d'exploiter%20les%20facteurs%20cl%C3%A9s.%3C%2FSPAN%3E%3C%2FP%3E%0A%3CP%3E%26gt%3B%26gt%3B%3CA%20href%3D%22https%3A%2F%2Fwww.jmp.com%2Fzh_tw%2Fevents%2Flive-webinars%2Fnon-series%2F2021-11-26.html%3Futm_source%3Djmp%26amp%3Butm_medium%3Dcommunity%26amp%3Butm_campaign%3Dwcl7015b00000578E6AAI%22%20target%3D%22_blank%22%20rel%3D%22noopener%20noreferrer%22%3E%20S'inscrire%20maintenant%3C%2FA%3E%3C%2FP%3E%0A%3CP%3E%3CA%20title%3D%22Croissance%20bas%C3%A9e%20sur%20les%20donn%C3%A9es%26nbsp%3B%3A%20faites%20bon%20usage%20de%20JMP%20pour%20passer%20%C3%A0%20une%20production%20au%20plus%20juste%20et%20%C3%A0%20une%20fabrication%20agile%22%20href%3D%22https%3A%2F%2Fwww.jmp.com%2Fzh_tw%2Fevents%2Flive-webinars%2Fnon-series%2F2021-11-26.html%3Futm_source%3Djmp%26amp%3Butm_medium%3Dcommunity%26amp%3Butm_campaign%3Dwcl7015b00000578E6AAI%22%20target%3D%22_blank%22%20rel%3D%22noopener%20noreferrer%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%20lia-image-align-left%22%20image-alt%3D%221126%20seminar-%20FB%20Ads-materials%20(2).png%22%20style%3D%22width%3A%20562px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%221126%20seminar-%20FB%20Ads-materials%20(2).png%22%20style%3D%22width%3A%20562px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%221126%20seminar-%20FB%20Ads-materials%20(2).png%22%20style%3D%22width%3A%20562px%3B%22%3E%3CSPAN%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%221126%20seminar-%20FB%20Ads-materials%20(2).png%22%20style%3D%22width%3A%20562px%3B%22%3E%3Cspan%20class%3D%22lia-inline-image-display-wrapper%22%20image-alt%3D%221126%20seminar-%20FB%20Ads-materials%20(2).png%22%20style%3D%22width%3A%20562px%3B%22%3E%3Cimg%20src%3D%22https%3A%2F%2Fcommunity.jmp.com%2Ft5%2Fimage%2Fserverpage%2Fimage-id%2F37362i9D8A10A1E5926DB9%2Fimage-dimensions%2F562x281%3Fv%3Dv2%22%20width%3D%22562%22%20height%3D%22281%22%20role%3D%22button%22%20title%3D%221126%20seminar-%20FB%20Ads-materials%20(2).png%22%20alt%3D%221126%20seminar-%20FB%20Ads-materials%20(2).png%22%20%2F%3E%3C%2Fspan%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FSPAN%3E%3C%2FA%3E%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3E%26nbsp%3B%3C%2FP%3E%3C%2FLINGO-BODY%3E%3CLINGO-TEASER%20id%3D%22lingo-teaser-433903%22%20slang%3D%22zh-TW%22%3E%3CP%3ELorsque%20vous%20analysez%20des%20donn%C3%A9es%2C%20quelle%20est%20la%20premi%C3%A8re%20%C3%A9tape%26nbsp%3B%3F%3C%2FP%3E%0A%3CP%3Eutilise%20le%20puissant%20et%20interactif%20g%C3%A9n%C3%A9rateur%20de%20graphiques%20JMP%20(%3CSPAN%3E%20g%C3%A9n%C3%A9rateur%20graphique%3C%2FSPAN%3E%20)%3CSPAN%3E%20Faire%20une%20analyse%20visuelle%20%3F%3C%2FSPAN%3E%3C%2FP%3E%0A%3CP%3Eou%20utilisez%20JMP%3CSPAN%3E%20de%3C%2FSPAN%3E%20Distribution%20(%3CSPAN%3E%20distribu%C3%A9%3C%2FSPAN%3E%20)%3CSPAN%3E%20La%20plate-forme%20effectue-t-elle%20une%20analyse%20des%20statistiques%20narratives%26nbsp%3B%3F%3C%2FSPAN%3E%3C%2FP%3E%0A%3CP%3EOu%20utilisez%20JMP%3CSPAN%3E%20Diff%C3%A9rents%20outils%20de%20mod%C3%A9lisation%20pr%C3%A9dictive%2C%20pour%20trouver%20le%20mod%C3%A8le%20pr%C3%A9dictif%20le%20plus%20adapt%C3%A9%20%3F%3C%2FSPAN%3E%3C%2FP%3E%0A%3CP%3EBien%20qu'il%20s'agisse%20du%20c%C3%A9l%C3%A8bre%20JMP%3CSPAN%3E%20fonction%2C%20mais%20comme%20utilisation%3C%2FSPAN%3E%20JMP%3CSPAN%3E%20Comme%20premi%C3%A8re%20%C3%A9tape%20de%20votre%20analyse%2C%20en%20plus%20des%20plates-formes%20ci-dessus%2C%20vous%20voudrez%20peut-%C3%AAtre%20envisager%20d'utiliser%3C%2FSPAN%3E%20Explorer%20les%20valeurs%20aberrantes%20(%3CSPAN%3E%20Explorer%20les%20valeurs%20aberrantes%3C%2FSPAN%3E%20)%3CSPAN%3E%20et%3C%2FSPAN%3E%20Explorer%20les%20valeurs%20manquantes%20(%3CSPAN%3E%20Explorer%20les%20valeurs%20manquantes%3C%2FSPAN%3E%20)%3CSPAN%3E%20Plate-forme%2C%20nettoyez%20et%20organisez%20vos%20informations.%3C%2FSPAN%3E%3C%2FP%3E%3C%2FLINGO-TEASER%3E
Choose Language Hide Translation Bar
運用JMP淨化離群值,以半導體分析WAT資料為例

當您分析資料時,第一個步驟是什麼?

是使用能力強大的互動性 JMP Graph  builder (圖形生成器) 做視覺化分析?

或是利用 JMP的Distribution (分佈) 平台做敘述統計量分析?

又或是利用 JMP 多樣的預測建模工具,找到最適當的預測模型?

 

雖然這些都是為人所樂道的 JMP 功能,但是做為使用 JMP 分析的第一步,除了上述的平台外,您不妨考慮利用 Explore Outliers (探索離群值)及 Explore Missing Values (探索遺失值) 平台,清潔整理您的資料。  

也許您認為探索遺失值還能理解,畢竟空值要先剃除對後續分析比較合理。但是,為什麼需要將離群值也排除?假如這樣的離群值正是代表資料母體的狀況,是否對我們的分析會有影響?而且,離群值很難抓取出來,不能先忽略這個影響嗎?

 

要釐清這些問題,我們需要先知道甚麼是離群值。

什麼是離群值?

我們先討論一維的狀況,隨機生成1000個標準常態分配Nor(0, 1)的值,如圖1,其中標記紅色的點,為超出[-4,4]範圍的離群點,這樣的發生機率為0.0063%,是非常小的機率,於是這這樣小機率區域上發生的點我們就視為離群值。

Michelle_Wu_0-1636341617235.png

 

圖1  隨機生成的常態分配點圖

 

接著我們來看二變數的狀況,利用 Graph builder 分析,不論資料點是兩個變數之間具有相同均值及變異數(X2 vs. X1)或是具有不同均值及變異數(X3 vs. X1),如下方圖二,我們都能用肉眼發現這些離群點(紅點)似乎跟其他的點有不一樣的趨勢,且與資料中心點的距離較遠。

 

Michelle_Wu_1-1636341617245.pngMichelle_Wu_2-1636341617252.png

圖2 (左) 2變數具有相同均值及變異數散點圖                      (右) 2變數具有不同均值及變異數散點圖

 

而當兩變數間有強相關性,也能發現有些離群點距離其他的樣本點之間的距離較大 (圖3 紅點)。另外,利用Fit Y by X平台的histogram boarder功能觀察兩變數的分配 (圖3),我們發現如果我們只關注單一變數,有些離群值我們會偵測不到,如下圖所示,這也是一個很好例子說明,當我們通盤考慮多變量的離群值,會抓到一些潛在的離群點,避免只觀察管控單一變量的誤判。幸運的,JMP提供了這樣的多變量離群點觀測平台。

 

Michelle_Wu_3-1636341617255.png

圖3 兩變數間有強相關性的散點圖

 

到此,我們能夠了解離群值的概念,就是找出相較於其他的樣本點,距離樣本中心較遠,或是點的座落位置與樣本整體「趨勢」不符合的點。有趣的是,離群點不代表有問題的點,而是代表與其他的樣本有著差距而無相同的「趨勢」。所以離群值,有可能代表良善社會的問題人物,也可能代表萬惡城市中的一股清流,就看你的樣本來自哪裡。

 

善用JMP分析WAT資料,解決半導體製程問題

說到這裡,我們用一個半導體製造的電性測試資料作為說明。

 

半導體晶圓製造的元件參數都有其對應電子特性,因此透過電性測試這些參數的結果便可反推對應製程的品質。晶圓代工廠一般稱電性參數測試稱為WAT (Wafer Acceptance Test),其客戶需要檢視WAT測試數據來確認製造品質。而新製程研發階段更需要蒐集大量WAT資料來解決製程問題或尋找最佳化的因子配置。

 

我們利用一組簡單的WAT資料作為示範,其中包含10個觀察參數資料,當我們直接分析產線三的資料,利用Analyze>Distribution 功能我們可以做出相應的直方圖及Outlier box plot 觀察是否有離群值,如圖四(左),Outlier Box 上的紅色線段標記出最集中涵蓋50%的區域,從此處我們可以發現資料集中在中間的區段。同樣的你也可以利用Graph Builder 的直方圖功能完成直方圖的繪製,如圖四(右)。

 

Michelle_Wu_4-1636341617257.png      Michelle_Wu_5-1636341617260.png

圖四(左) Distribution 平台的直方圖;Outlier box plot (右) Graph builder的Box plot

 

根據Quantile Range 離群值檢驗我們發現參數一存在一個離群值(紅點),利用JMP的Analyze>Screening>Explore Outliers 平台中的Quantile Range Outliers 功能可以發現其離群值(圖五),並可以針對資料做標記、排除、視為遺失值等動作。

 

同時你也可以調整判定為離群值的規則,可以調整尾端百分位值(Tail Quantile)和其倍數(Q),以常見的IQR離群值檢驗,上限為Q3+1.5*IQR,上限為Q1-1.5*IQR為例, Tail Quantile 即為0.25, Q則為1.5。而當您的資料較不符合常態分配時,可以考慮使用Robust Fit Outliers作為離群值篩選平台。

 

Michelle_Wu_6-1636341617263.png

圖五 Explore Outliers 平台中的Quantile Range Outliers

 

但當我們更全面考慮,納入所有產線的資料,你會發現產線三擁有較高的參數一測試結果,如圖六,原本產線三上的離群值8.01反而較符合大多數資料的測試結果,而實際的離群值則轉為產線三最高的三個值 (圖六紅點),這樣的結果也呼應前述樣本出處的重要性,廣泛獲取能夠代表母體資料對於篩選正確的離群值有舉足輕重的腳色。

 

Michelle_Wu_7-1636341617265.png

圖六 全部產線的Box plot

 

當考慮全部參數的影響,找出對於所有參數影響下的離群值時,JMP也提供Robust PCA Outliers 或是K Nearest Neighbor Outliers 功能供您使用。您可以在Analyze> Screening> Explore Outliers 下找到這兩個功能平台,這邊利用K Nearest Neighbor Outliers 平台找尋離群值,利用複合選取距離較高的樣本點,標記為離群值(圖七)。

 

Michelle_Wu_8-1636341617275.png

圖七 K Nearest Neighbor Outliers 平台下的距離圖

 

藉由Graph builder 確認這些被選為離群值的資料點在每一個參數上的分佈位置可以發現不是所有的點都屬於單一參數的離群值(如資料點290, 295, 296, 297, 298, 299),這樣的結果也和前述提及藉由JMP的多變量離群值篩選平台,可以抓出許多在觀測單變量下所忽略的潛在問題點(圖八)。

 

Michelle_Wu_9-1636341617283.png

圖八 用Boxplot 比較離群值分別在變數上的分佈

 

移除離群值,建構更準確的預測模型

我們知道了JMP有很好的平台可以幫助我們把離群值篩選出來,但是,沒有把離群值篩選掉會造成甚麼影響? 化繁為簡,我們先看一下兩參數間的離群值影響。

 

我們觀測WAT資料參數一級參數二之間的2次效應迴歸式估計的R-square值,在還沒移除篩選出的離群值前,R-square值為0.602,如圖九(a),而移除離群值後,R-square值上升為0.740,如圖九(b),這說明移除離群值能夠建構出更準確的模型,擁有差距更小的估計值。

 

當我們利用這些參數建構出預測結果,例如良率預估或是CP/FT 測試預估值等,我們能更有效偵測不良,降低成本,增加效益。

 

Michelle_Wu_10-1636341617287.pngMichelle_Wu_11-1636341617291.png

             圖九(a) 沒移除離群值的二次迴歸估計                              (b)移除離群值的二次迴歸估計

 

接著放大考量全部參數,並用這些參數預估後續Bin值結果,找到關鍵因子。並在最小化Bin值的設定下找到最佳因子設定值。相同的,我們先考量不移除離群值,找出最適合的迴歸模型,利用Fit model平台,我們得到迴歸模型R-square 大約為0.971,包含許多因子效應項,這其中可以看到許多的交互作用項效應比主效應強, 似乎比較混亂,且觀察Residual by Row Plot 可以發現篩選為離群值的那幾個樣本差距是比較大的(圖十)。

 

Michelle_Wu_12-1636341617296.pngMichelle_Wu_13-1636341617307.png

圖十  不移除離群值迴歸式R-square、Residual by Row Plot及影響的因子效應項

 

下一步把離群值移除,再做一次迴歸估計,我們發現,不僅R-square 上升到0.999,關鍵因子也縮減到四個因子效應項,觀察Residual by Row Plot,也沒有發現差異比較大的樣本(圖十一)。根據此迴歸公式預測的結果也會更準確,避免做出錯誤的判斷。

 

Michelle_Wu_14-1636341617309.png

Michelle_Wu_15-1636341617310.png

圖十一   不移除離群值迴歸式R-square、Residual by Row Plot及影響的因子效應項

 

蒐集多一點資料做後續測試迴歸式的準確度,我們可以在Analyze>Fit Y by X的平台下,比較迴歸式的預估值及實際的Bin 值,新資料依舊能夠有很好的水準。R-square有0.999的準確度,而且沒有特殊的點有過大的差異。

 

Michelle_Wu_16-1636341617313.png

圖十二   比較迴歸式的預估值及實際的Bin 值

 

藉由JMP的Explore Outliers平台,我們能夠輕易快速的排除離群值,避免受到離群值影響誤判因子之間相關性或是做出錯誤的模型預測,導致錯誤的預估及決策。作為分析的前哨站,JMP的Explore Outliers 平台能夠針對不同情景,不論單變數或是多變數,資料是否為常態,都能提出適當的平台幫助使用者找到離群值,快速達到資料清理的目的

 

衍伸閱讀:

如何在30秒內在JMP中製作晶圓圖(英)

使用功能數據分析進行圖像和晶圓分析

 

報名11/26半導體研討會

>> 透過 JMP如何讓使用者快速有效的得到最佳化的參數組合,進而提昇產線良率及增加產出?報名11/26半導體研討會,瞭解如何善用 JMP 進行WAT分析,改善製程良率及挖掘關鍵因子。

>> 立即報名

1126 seminar- FB Ads-materials (2).png

 

 

Last Modified: Nov 8, 2021 8:48 AM