Ukuhlaziywa Kwe-Multivariate Kocwaningo Lwezimboni
Ezweni lezimboni eliyinkimbinkimbi ngokwengeziwe, izinqumo eziqhutshwa idatha ziba yisidingo esingaxoxiswana ngaso. Izinkampani ezikhiqizayo, ezamandla, zamakhemikhali, zemithi, zezokuthutha, ngisho nezinsizakalo ezisekelwe kubuchwepheshe zibhekene nenani elikhulu lezinto eziguquguqukayo: ikhwalithi yezinto zokusetshenziswa, izilungiselelo zomshini, izinga lokushisa lenqubo, ingcindezi, isikhathi somjikelezo, izimo zomsebenzisi, ukwakheka komkhiqizo, isidingo semakethe, kanye nezici zemvelo. Lapho lezi zinto eziguquguqukayo zihlangana, ukuhlaziya idatha ngayinye (i-univariate) noma izinto ezimbili kuphela eziguquguqukayo (i-bivariate) ngokuvamile akwanele ukuqonda imbangela yezinkinga noma ukwakha amasu okuthuthukisa. Yilapho ukuhlaziywa kwe-multivariate kudlala khona indima ebalulekile: ukusiza abacwaningi bemboni baqonde ubudlelwano obuhambisanayo phakathi kwezinto eziningi eziguquguqukayo, ukuthola amaphethini afihliwe, nokwakha amamodeli okubikezela anembe kakhudlwana.
Kuyini Ukuhlaziywa Kwe-Multivariate?
Ukuhlaziywa kwe-Multivariate kuyisethi yezindlela zokufunda zezibalo nezomshini ezisetshenziswa ukuhlaziya idatha eneziguquguquko eziningi ngasikhathi sinye. Izinhloso zingahluka, njengokuhlanganisa izinto ngokusekelwe ekufaneni, ukunciphisa ubukhulu bedatha ukuze kube nokufingqa okukhulu, ukuhlola umehluko phakathi kwamaqembu anezimpendulo eziningi, noma ukubikezela umphumela wenqubo kusuka kusethi yokufaka. Esimweni sezimboni, "izinto" zingasho imikhiqizo, amaqoqo okukhiqiza, imishini, abaphakeli, amakhasimende, noma ngisho nokushintshana komsebenzi. Iziguquguquko zingafaka izilinganiso zomzimba, imingcele yenqubo, izinkomba zekhwalithi, ngisho nezilinganiso zezimali.
Ngokungafani nezindlela ezilula ezihlola i-variable eyodwa ngesikhathi, ukuhlaziywa kwe-multivariate kubheka uhlelo lonke. Lokhu kubalulekile ngoba ukwehluleka kwekhwalithi noma ukuwohloka kokusebenza akuvamile ukubangelwa yisici esisodwa. Isibonelo, amaphutha omkhiqizo angavela ekuhlanganisweni komswakama wezinto zokusetshenziswa, izinga lokushisa lokushisa, isivinini somshini, kanye nekhono lomsebenzisi. Ukuhlaziywa kwe-multivariate kusiza ukuthola lezi zinhlanganisela ngobuningi.
Kungani Kubalulekile Ocwaningweni Lwezimboni?
Ucwaningo lwezimboni luvame ukudinga ukusebenza kahle kwezindleko, ukufika ngesikhathi, kanye nokufaneleka okusebenzayo. Izindlela eziningi zinikeza izinzuzo eziningana ezibalulekile:
1. Ukuqonda ukusebenzisana okuguquguqukayo: Izinqubo eziningi zezimboni ziyaxhomekeka komunye nomunye. Ukuhlaziywa kwe-multivariate kungabamba ukuhlobana kanye nokusebenzisana okuthinta umkhiqizo.
2. Ukunciphisa ubunzima: Idatha yezimboni ivame ukuba nobukhulu obuphezulu (amashumi kuya kwamakhulu eziguquguquko zenzwa). Amasu okunciphisa ubukhulu angenza idatha ibe lula ngaphandle kokulahla ulwazi olubalulekile.
3. Ukutholwa kwe-Anomaly nokulawulwa kwekhwalithi: Amamodeli e-multivariate angabona amaphethini "avamile" futhi anikeze izixwayiso lapho izinqubo ziphambuka.
4. Ukuhlukaniswa kwemakethe/umkhiqizo: Ukuqoqa amakhasimende noma imikhiqizo ukuze kutholakale amasu okumaketha afanele, ukuklama, kanye nokuphathwa kwephothifoliyo.
5. Ukubikezela nokwenza ngcono: Ukubikezela ukwehluleka komshini, isidingo, noma ikhwalithi yomkhiqizo ukuze kusekelwe ukulungiswa nokuhlela ukukhiqiza.
Izinhlobo Zezindlela Eziningi Ezisetshenziswa Kakhulu Embonini
1. Ukuhlaziywa Kwezingxenye Eziyinhloko (i-PCA)
I-PCA (Ukuhlaziywa Kwezingxenye Eziyinhloko) isetshenziselwa ukunciphisa ubukhulu bedatha ngokuguqula iziguquguquko zokuqala zibe yinani lezingxenye ezintsha, ezizimele (eziqondile). Embonini, i-PCA ivame ukusetshenziswa uku:
- hlanganisa iziguquguquko zenzwa zibe izinkomba eziningana eziyinhloko,
- thola ukuzulazula kwenqubo,
– cabanga ngomehluko phakathi kwamaqoqo noma phakathi kwemigqa yokukhiqiza.
Isibonelo, ifektri yokudla ingaba namarekhodi amaningi okushisa, umswakama, kanye nesikhathi sokupheka. I-PCA ingabonisa ukuthi ukwehluka okukhulu kwekhwalithi kuchazwa yinhlanganisela ethile yokushisa kanye nomswakama, okuvumela ithimba lenqubo ukuthi ligxile kumapharamitha abalulekile.
2. Ukuhlaziywa Kwamaqoqo (Ukuhlanganisa)
Ukuqoqa amaqembu kusekelwe ekufaneni. Izindlela ezidumile zifaka phakathi i-K-Means, i-Hierarchical Clustering, kanye ne-DBSCAN. Izinhlelo zokusebenza zezimboni zifaka:
- ukuhlukaniswa kwamakhasimende okusekelwe ekuziphatheni kokuthenga,
– ukuhlanganisa abahlinzeki ngeqembu ngokusekelwe kwikhwalithi nesikhathi sokulethwa,
– thola amaphethini afanayo okwehluleka komshini.
Ngamaqoqo, izinkampani zingahlukanisa phakathi kwamakhasimende agxile emananini, amakhasimende e-premium, kanye namakhasimende “esikhathini esithile”, bese ziklama amasu ahlukene eqembu ngalinye.
3. Ukuhlaziywa Okuhlukile Nokuhlukaniswa
Ukuhlaziywa okuhlukile (isb., i-LDA) kanye namamodeli okuhlukaniswa (ukuhlehliswa kwempahla, ihlathi elingahleliwe, i-SVM) kusetshenziswa lapho umgomo wocwaningo uwukuhlukanisa izigaba ezithile: imikhiqizo "yokudlula" noma "yokwehluleka", imishini "enempilo" noma "esengozini", amakhasimende "a-churn" noma "athembekile". Ekukhiqizeni, amamodeli okuhlukaniswa angaxhumanisa amapharamitha enqubo namathuba okukhubazeka, okuvumela opharetha ukuthi benze izinguquko ngaphambi kokuba imikhiqizo yehluleke.
4. Amamodeli Okubuyela Emuva Kaningi kanye Nokubikezela
Izindlela eziningi zokuhlehla okuqondile, ukuhlehla kwe-ridge/lasso, kanye nezindlela ezingezona eziqondile ezifana nokukhulisa i-gradient zisetshenziselwa ukubikezela umphumela ngokusekelwe kokufakwayo okuningi. Izibonelo:
- bikezela isivuno sokukhiqiza kusukela ekwakhiweni kwezinto kanye nezilungiselelo zomshini,
- bikezela izikhathi zokulethwa ezivela emizileni, isimo sezulu, kanye nokuminyana,
– bikezela ukusetshenziswa kwamandla kusukela kumthwalo wenjini kanye nokushisa okuzungezile.
Izindlela zokulungisa kabusha (i-ridge/lasso) ziwusizo kakhulu lapho iziguquguquko ziningi futhi zihlobene, okuyisimo esivamile kudatha yezinzwa zezimboni.
5. Ukuhlaziywa kwe-MANOVA kanye ne-Multiple Response
Ekuhlolweni kwezimboni, kuvame ukuba nempendulo yekhwalithi engaphezu kweyodwa: amandla okudonsa, ubulukhuni, ukumelana nokushisa, kanye nesilinganiso sokukhubazeka. I-MANOVA (Ukuhlaziywa Kwe-Multivariate of Variance) isetshenziselwa ukuhlola ukuthi ukwelashwa okuthile (isb., uhlobo lwezinto noma indlela yokucubungula) kuthinta yini izimpendulo eziningi ngasikhathi sinye. Lokhu kusebenza kahle kakhulu kunokuhlola ngayinye ngayinye futhi kunciphisa nengozi yephutha lezibalo ngenxa yokuhlolwa okuphindaphindiwe.
Izigaba Zokusebenzisa Ukuhlaziywa Kwezinto Eziningi Ocwaningweni Lwezimboni
Ukuze imiphumela isebenze futhi isebenze, uhlelo lokusebenza lwe-multivariate kufanele lulandele ukugeleza okuhlelekile:
1. Ukwakhiwa kwezinkinga nezinhloso
Nquma ukuthi ukugxila kuwukuhlola amaphethini, ukuhlukaniswa kwezigaba, ukuhlolwa kwe-hypothesis, noma ukubikezela. Le nhloso inquma indlela esetshenziswayo.
2. Ukuqoqwa kwedatha nokuqonda
Idatha yezimboni ingavela ezinzwa ze-ERP, SCADA, IoT, ukuhlolwa kwekhwalithi, noma izinhlolovo zamakhasimende. Kubalulekile ukuqonda incazelo yenguquko ngayinye, imvamisa yokurekhoda, kanye nethuba lokukhetha.
3. Ukuhlanzwa kwedatha
Phatha idatha engekho, izinto ezingaphandle, umehluko wesikali, kanye nokuphindaphinda. Izindlela eziningi ze-multivariate zizwela isikali, ngakho-ke ukulinganisa (i-z-score) kuvame ukudingeka.
4. Ukuhlola kokuqala
Ukubona ngeso lengqondo ukuhlobana, ukusatshalaliswa, kanye ne-scatter matrix kusiza ukudweba ubudlelwano obuyisisekelo ngaphambi kokumodela kwe-multivariate.
5. Ukumodela nokuqinisekisa
Sebenzisa amasu okuqinisekisa njengokuqinisekisa okuphambene, amasethi okubamba, noma ukufaka i-bootstrapping. Embonini, ukuqinisekisa kubalulekile ngoba imodeli izothinta izinqumo zokusebenza.
6. Ukuhunyushwa kanye nokusetshenziswa
Imodeli kumele ihunyushwe ibe yizenzo: ukuthi yiziphi izinhlaka zenqubo okufanele zisethwe, yiziphi izingxenye okufanele ziqondiswe, noma yiziphi i-alamu okufanele zisethwe ohlelweni lokuqapha.
7. Ukuqapha okuqhubekayo
Izinqubo zezimboni ziyashintsha (imishini iyaguga, abaphakeli bayashintsha, isidingo siyashintsha). Amamodeli amaningi kumele ahlolwe kabusha ngezikhathi ezithile ukuze kugwenywe ukwehla kokunemba.
Izinselele Ezivamile Nendlela Yokuzinqoba
Ucwaningo lwezimboni lubhekene nezinselele ezihlukile. Okokuqala, i-multicollinearity, lapho iziguquguquko zokufaka zihlobene kakhulu. Izixazululo zingafaka phakathi i-PCA, ukuhlaziywa kwe-ridge/lasso, noma ukukhetha iziguquguquko ezifanele kakhulu. Okwesibili, idatha engalingani, njengalapho amaphutha ebalelwa ku-1% kuphela womkhiqizo. Lokhu kungaxazululwa ngamasu okulinganisa (ukwenza isampula ngaphezulu/ukwenza isampula ngaphansi) kanye nezilinganiso zokuhlola ezifanele (ukukhumbula ngokunemba, i-F1, i-AUC). Okwesithathu, ikhwalithi yedatha: izinzwa zingaba nephutha, futhi ukuqoshwa ngesandla kungangaguquki. Ukutshala imali ekuphathweni kwedatha nasekulinganisweni kwezinzwa kuvame ukubaluleka njengokukhethwa kwe-algorithm.
Ngaphezu kwalokho, kunenkinga yokuhunyushwa kalula. Imboni ivame ukudinga izincazelo ezicacile ukuqinisekisa ukuthi izincomo ziyamukelwa ngabaphathi kanye nabaqhubi. Uma imodeli iyinkimbinkimbi (isb., amanethiwekhi okukhulisa noma amanethiwekhi e-neural), sebenzisa izindlela zokuhumusha ezifana nokubaluleka kwesici, ama-partial dependence plots (PDP), noma i-SHAP ukuchaza iminikelo eguquguqukayo.
Isiphetho
Ukuhlaziywa kwe-multivariate kuyisisekelo esibalulekile ocwaningweni lwezimboni lwanamuhla ngoba lubamba ubunzima bezinhlelo zomhlaba wangempela ezithonywa izici eziningi. Besebenzisa izindlela ezifana ne-PCA, ukuqoqa, ukuhlukaniswa, ukuhlehliswa kwe-multivariate, kanye ne-MANOVA, abacwaningi bangathola amaphethini, bahlole imibono, babikezele ikhwalithi, futhi bathuthukise izinqubo zokukhiqiza nezebhizinisi. Izihluthulelo zempumelelo zisezinhlosweni zocwaningo ezicacile, ikhwalithi enhle yedatha, ukukhethwa kwezindlela ezifanele, ukuqinisekiswa okuqinile, kanye nekhono lokuhumusha imiphumela yezibalo ibe yizinqumo zokusebenza. Uma kusetshenziswa kahle, ukuhlaziywa kwe-multivariate akuyona nje ithuluzi lokuhlaziya kodwa ithuluzi eliyisu lokuthuthukisa ukusebenza kahle kwezimboni, ikhwalithi, kanye nokuncintisana.
Uma ufisa, ngingahlela lesi sihloko ngokwemboni ethile (isib., ukukhiqiza, imithi, ezokuthutha, noma amandla), ngifake izibonelo zesifundo secala, noma ngifake uhlaka oluphelele lwendlela yocwaningo (izinhloso, iziguquguquko, amathuluzi, kanye nomklamo wokuhlaziya).
