{"id":30044,"date":"2019-10-31T21:33:22","date_gmt":"2019-10-31T18:33:22","guid":{"rendered":"https:\/\/prohoster.info\/blog\/kak-my-predskazyvali-ottok-podojdya-k-nemu-kak-k-stihijnomu-bedstviyu\/"},"modified":"2019-10-31T21:33:22","modified_gmt":"2019-10-31T18:33:22","slug":"kak-my-predskazyvali-ottok-podojdya-k-nemu-kak-k-stihijnomu-bedstviyu","status":"publish","type":"post","link":"https:\/\/prohoster.info\/sk\/blog\/kak-my-predskazyvali-ottok-podojdya-k-nemu-kak-k-stihijnomu-bedstviyu","title":{"rendered":"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"<p>Niekedy, aby vyrie\u0161ili nejak\u00fd probl\u00e9m, je potrebn\u00e9 sa na\u0148 pozrie\u0165 z in\u00e9ho uhla. Aj ke\u010f sa v posledn\u00fdch desiatich rokoch podobn\u00e9 probl\u00e9my rie\u0161ili rovnak\u00fdm sp\u00f4sobom s r\u00f4znymi \u00fa\u010dinkami, nie je ist\u00e9, \u017ee tento sp\u00f4sob je jedin\u00fd.<\/p>\n<p>Existuje t\u00e9ma, ako odtok klientov. Je to nevyhnutn\u00e9, preto\u017ee klienti akejko\u013evek spolo\u010dnosti m\u00f4\u017eu z mnoh\u00fdch d\u00f4vodov presta\u0165 vyu\u017e\u00edva\u0165 jej produkty alebo slu\u017eby. Pre spolo\u010dnos\u0165 je odtok - aj ke\u010f prirodzen\u00fd - nie naj\u017eiadanej\u0161\u00ed jav, a preto sa v\u0161etci sna\u017eia tento odtok minimalizova\u0165. E\u0161te lep\u0161ie je predpoveda\u0165 pravdepodobnos\u0165 odtoku ur\u010ditej kateg\u00f3rie pou\u017e\u00edvate\u013eov alebo konkr\u00e9tneho pou\u017e\u00edvate\u013ea a navrhn\u00fa\u0165 nejak\u00e9 kroky na udr\u017eanie.<\/p>\n<p>Analyzova\u0165 a pok\u00fasi\u0165 sa zadr\u017ea\u0165 klienta, ak je to mo\u017en\u00e9, je potrebn\u00e9 minim\u00e1lne z t\u00fdchto d\u00f4vodov:<\/p>\n<ul>\n<li><b>z\u00edskanie nov\u00fdch klientov je drah\u0161ie ne\u017e procesy udr\u017eania<\/b>. Na z\u00edskanie nov\u00fdch klientov je zvy\u010dajne potrebn\u00e9 vynalo\u017ei\u0165 ur\u010dit\u00e9 prostriedky (reklama), zatia\u013e \u010do existuj\u00facich klientov je mo\u017en\u00e9 aktivova\u0165 \u0161peci\u00e1lnou ponukou s osobitn\u00fdmi podmienkami;<\/li>\n<li><b>pochopenie d\u00f4vodov odchodu klientov je k\u013e\u00fa\u010dom k zlep\u0161eniu produktov a slu\u017eieb<\/b>.<\/li>\n<\/ul>\n<p>\nExistuj\u00fa \u0161tandardn\u00e9 pr\u00edstupy k predpovedaniu odtoku. No na jednom z majstrovstiev v AI sme sa rozhodli vysk\u00fa\u0161a\u0165 na tento \u00fa\u010del Weibullovo rozdelenie. \u010casto sa pou\u017e\u00edva na anal\u00fdzu pre\u017eitia, predpovedanie po\u010dasia, anal\u00fdzu pr\u00edrodn\u00fdch katastrof, v priemyselnej in\u017einierstve a podobne. Weibullovo rozdelenie je \u0161peci\u00e1lna distribu\u013ea\u010dn\u00e1 funkcia, parametrizovan\u00e1 dvoma parametrami. <img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/d30a25becc44269d51957349c6e8b4a0.png\" style=\"display:block;margin: 0 auto;\" \/> a <img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/42092030604bf54a98c8379fd532bffe.png\" style=\"display:block;margin: 0 auto;\" \/>. <\/p>\n<p><noindex><a rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/company\/rostelecom\/blog\/444694\/\"><img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/62e1933c8335dd16d765f0e358952d84.png\" style=\"display:block;margin: 0 auto;\" \/><\/a><\/noindex><br \/>\n<i>Wikipedia<\/i><\/p>\n<p>Celkovo, zauj\u00edmav\u00e1 t\u00e9ma, ale na predpovedanie odtoku a v\u0161eobecne vo fintechu sa nepou\u017e\u00edva tak \u010dasto. Pod katom sa dozviete, ako sme to (Laborat\u00f3rium inteligentnej anal\u00fdzy d\u00e1t) spravili, pri\u010dom sme z\u00edskali zlato na Majstrovstv\u00e1ch v umel\u00fa inteligenciu v kateg\u00f3rii \u201eAI v bank\u00e1ch.\u201c<br \/>\n<noindex><a rel=\"nofollow\" name=\"habracut\"><\/a><\/noindex><\/p>\n<h2>O odtoku vo v\u0161eobecnosti<\/h2>\n<p>\nPoj\u010fme si trochu objasnit, co je to odliv z\u00e1kazn\u00edk\u016f a pro\u010d je tak d\u016fle\u017eit\u00fd. Pro podnik\u00e1n\u00ed je d\u016fle\u017eit\u00e1 z\u00e1kaznick\u00e1 z\u00e1kladna. Do t\u00e9to z\u00e1kladny p\u0159ich\u00e1zej\u00ed nov\u00ed z\u00e1kazn\u00edci, nap\u0159\u00edklad kdy\u017e se dozv\u011bd\u00ed o produktu nebo slu\u017eb\u011b z reklamy, n\u011bjakou dobu existuj\u00ed (aktivn\u011b pou\u017e\u00edvaj\u00ed produkty) a po n\u011bjak\u00e9 dob\u011b p\u0159estanou pou\u017e\u00edvat. Tento \u010dasov\u00fd \u00fasek naz\u00fdv\u00e1me \u201e\u017divotn\u00ed cyklus z\u00e1kazn\u00edka\u201c (angl. Customer Lifecycle) \u2014 to je term\u00edn, kter\u00fd popisuje f\u00e1ze, kter\u00fdmi z\u00e1kazn\u00edk proch\u00e1z\u00ed, kdy\u017e se dozv\u00ed o produktu, rozhodne se k n\u00e1kupu, zaplat\u00ed, pou\u017e\u00edv\u00e1 a st\u00e1v\u00e1 se loaj\u00e1ln\u00edm spot\u0159ebitelem, a nakonec p\u0159estane pou\u017e\u00edvat z r\u016fzn\u00fdch d\u016fvod\u016f produkty. Odpov\u00eddaj\u00edc\u00edm zp\u016fsobem odliv je z\u00e1v\u011bre\u010dn\u00e1 f\u00e1ze \u017eivotn\u00edho cyklu z\u00e1kazn\u00edka, kdy z\u00e1kazn\u00edk p\u0159est\u00e1v\u00e1 vyu\u017e\u00edvat slu\u017eby, a pro podnik\u00e1n\u00ed to znamen\u00e1, \u017ee z\u00e1kazn\u00edk p\u0159estal p\u0159in\u00e1\u0161et zisk a jakoukoli jinou hodnotu. <\/p>\n<p>Ka\u017ed\u00fd bankovn\u00ed klient je konkr\u00e9tn\u00ed \u010dlov\u011bk, kter\u00fd si vybere ur\u010ditou bankovn\u00ed kartu p\u0159esn\u011b podle sv\u00fdch pot\u0159eb. \u010casto cestuje \u2014 hod\u00ed se karta s m\u00edlemi. Hodn\u011b nakupuje \u2014 zdrav\u00edm, karta s cashbackem. Hodn\u011b nakupuje ve specifick\u00fdch obchodech \u2014 a pro to u\u017e je tady speci\u00e1ln\u00ed partnersk\u00fd plast. Samoz\u0159ejm\u011b, n\u011bkdy se karta vyb\u00edr\u00e1 i podle krit\u00e9ria \u201eNejdra\u017e\u0161\u00ed \u00fadr\u017eba\u201c. Obecn\u011b je tu dost prom\u011bnn\u00fdch.<\/p>\n<p>A je\u0161t\u011b si \u010dlov\u011bk vyb\u00edr\u00e1 banku \u2014 jak\u00fd m\u00e1 smysl vyb\u00edrat si kartu banky, jej\u00ed\u017e pobo\u010dky jsou pouze v Moskv\u011b a okol\u00ed, kdy\u017e jsi z Chabarovsku? A\u0165 je karta takov\u00e9 banky jakkoli v\u00fdhodn\u00e1, p\u0159\u00edtomnost pobo\u010dek banky v bl\u00edzkosti st\u00e1le z\u016fst\u00e1v\u00e1 d\u016fle\u017eit\u00fdm krit\u00e9riem. Ano, 2019 je u\u017e tady a digit\u00e1ln\u00ed technologie \u2014 to je v\u0161e, ale n\u011bkter\u00e9 ot\u00e1zky u n\u011bkter\u00fdch bank lze vy\u0159e\u0161it pouze na pobo\u010dce. Plus, op\u011bt n\u011bjak\u00e1 \u010d\u00e1st populace mnohem v\u00edce d\u016fv\u011b\u0159uje fyzick\u00e9 bance, a ne aplikaci ve smartphonu, to je tak\u00e9 nutn\u00e9 vz\u00edt v \u00favahu.<\/p>\n<p>Jak vypl\u00fdv\u00e1, d\u016fvod\u016f pro odm\u00edtnut\u00ed bankovn\u00edch produkt\u016f (nebo samotn\u00e9 banky) m\u016f\u017ee b\u00fdt mnoho. Zm\u011bnil pr\u00e1ci, a tarif karty se zm\u011bnil z platov\u00e9ho na \u201ePro oby\u010dejn\u00e9 smrteln\u00edky\u201c, kter\u00fd je m\u00e9n\u011b v\u00fdhodn\u00fd. P\u0159est\u011bhoval se do jin\u00e9ho m\u011bsta, kde nejsou pobo\u010dky banky. Nezamlouvalo se mu jedn\u00e1n\u00ed s nekvalifikovan\u00fdm oper\u00e1torem na pobo\u010dce. To znamen\u00e1, \u017ee d\u016fvod\u016f pro uzav\u0159en\u00ed \u00fa\u010dtu m\u016f\u017ee b\u00fdt i mnohem v\u00edce ne\u017e pro pou\u017e\u00edv\u00e1n\u00ed produktu.<\/p>\n<p>Z\u00e1kazn\u00edk nemus\u00ed len jasne vyjadri\u0165 svoj \u00famysel - pr\u00eds\u0165 do banky a nap\u00edsa\u0165 \u017eiados\u0165, ale jednoducho m\u00f4\u017ee presta\u0165 pou\u017e\u00edva\u0165 produkty bez ukon\u010denia zmluvy. Preto bolo rozhodnut\u00e9 pou\u017ei\u0165 strojov\u00e9 u\u010denie a AI na pochopenie tak\u00fdchto \u00faloh.<\/p>\n<p>Okrem toho m\u00f4\u017ee odliv klientov nast\u00e1va\u0165 v akejko\u013evek oblasti (telekomunik\u00e1cie, internetov\u00ed poskytovatelia, pois\u0165ovne, skr\u00e1tka tam, kde existuje z\u00e1kazn\u00edcka z\u00e1klad\u0148a a periodick\u00e9 transakcie).<\/p>\n<h2>\u010co sme urobili<\/h2>\n<p>\nPredov\u0161etk\u00fdm bolo potrebn\u00e9 jasne vymedzi\u0165 hranicu - od kedy za\u010d\u00edname po\u010d\u00edta\u0165 klienta ako od\u00edden\u00e9ho. Z poh\u013eadu banky, ktor\u00e1 n\u00e1m poskytla \u00fadaje na spracovanie, bol stav aktivity klienta bin\u00e1rny - bu\u010f akt\u00edvny, alebo nie. V tabu\u013eke 'Aktivita' bol flag ACTIVE_FLAG, ktor\u00e9ho hodnota mohla by\u0165 bu\u010f '0' alebo '1' (pod\u013ea toho 'Neakt\u00edvny' a 'Akt\u00edvny'). A v\u0161etko by bolo v poriadku, ale \u010dlovek m\u00e1 tendenciu nejak\u00fd \u010das akt\u00edvne pou\u017e\u00edva\u0165 slu\u017eby a potom na mesiac vypadn\u00fa\u0165 z akt\u00edvnych - ochorel, i\u0161iel do inej krajiny na dovolenku, alebo dokonca za\u010dal testova\u0165 kartu inej banky. Po dlh\u0161ej dobe neaktivity v\u0161ak m\u00f4\u017ee op\u00e4\u0165 za\u010da\u0165 vyu\u017e\u00edva\u0165 slu\u017eby banky.<\/p>\n<p>Preto sme sa rozhodli naz\u00fdva\u0165 obdob\u00edm neaktivity ur\u010dit\u00fd nep\u0159eru\u0161en\u00fd \u010dasov\u00fd \u00fasek, po\u010das ktor\u00e9ho bol pre neho flag nastaven\u00fd ako '0'. <\/p>\n<p><img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/c4263f11be0f122e54ed7985ba0d2c9d.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<br \/>\nKlienti prech\u00e1dzaj\u00fa z neakt\u00edvnych na akt\u00edvnych po r\u00f4znych d\u013a\u017ekach obdob\u00ed neaktivity. M\u00e1me mo\u017enos\u0165 vypo\u010d\u00edta\u0165 empirick\u00fa veli\u010dinu 'spo\u013eahlivos\u0165 obdob\u00ed neaktivity' - teda pravdepodobnos\u0165, \u017ee osoba op\u00e4\u0165 za\u010dne pou\u017e\u00edva\u0165 produkty banky po do\u010dasnej neaktivite. <\/p>\n<p>Napr\u00edklad, na tomto grafe je zn\u00e1zornen\u00e9 obnovenie aktivity (ACTIVE_FLAG=1) klientov po nieko\u013ek\u00fdch mesiacoch neaktivity (ACTIVE_FLAG=0).<\/p>\n<p><img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/3828c06a86ca6b43e77a6de61f768804.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<br \/>\nTu trochu objasn\u00edme s\u00fabor \u00fadajov, s ktor\u00fdmi sme za\u010dali pracova\u0165. Banka poskytla agregovan\u00e9 inform\u00e1cie za 19 mesiacov v nasleduj\u00facich tabu\u013ek\u00e1ch:<\/p>\n<ul>\n<li>'Aktivita' - mesa\u010dn\u00e9 transakcie klientov (pod\u013ea kariet, v internetovom bankovn\u00edctve a mobilnej aplik\u00e1cii), vr\u00e1tane v\u00fdplat a inform\u00e1ci\u00ed o obrate.<\/li>\n<li>'Karty' - \u00fadaje o v\u0161etk\u00fdch kart\u00e1ch, ktor\u00e9 m\u00e1 klient, s podrobnou tarifnou sie\u0165ou.<\/li>\n<li>\u201eZmluvy\u201c \u2013 inform\u00e1cie o zmluv\u00e1ch klienta (ako otvoren\u00fdch, tak aj uzavret\u00fdch): \u00favery, vklady a in\u00e9, so uveden\u00edm parametrov ka\u017ed\u00e9ho z nich.<\/li>\n<li>\u201eKlienti\u201c \u2013 s\u00fabor demografick\u00fdch \u00fadajov (pohlavie a vek) a dostupnos\u0165 kontaktn\u00fdch \u00fadajov.<\/li>\n<\/ul>\n<p>\nNa na\u0161u pr\u00e1cu sme potrebovali v\u0161etky tabu\u013eky, okrem \u201eKariet\u201c. <\/p>\n<p>Zlo\u017eitost tu spo\u010d\u00edvala aj v tom, \u017ee banka v t\u00fdchto \u00fadajoch neuv\u00e1dzala, ak\u00e1 presne aktivita prebiehala na kart\u00e1ch. Teda pochopi\u0165, \u010di prebehli transakcie alebo nie, sme dok\u00e1zali, ale ur\u010di\u0165 ich typ - u\u017e nie. Preto nebolo jasn\u00e9, \u010di klient vyberal hotovos\u0165, dost\u00e1val mzdu, alebo utr\u00e1cal peniaze na n\u00e1kupy. A navy\u0161e sme nemali \u00fadaje o zostatkoch na \u00fa\u010dtoch, \u010do by bolo u\u017eito\u010dn\u00e9. <\/p>\n<p>Samotn\u00fd v\u00fdber bol nesliepan\u00fd - v tomto reze za 19 mesiacov banka neuplatnila \u017eiadne pokusy o udr\u017eanie klientov a minimaliz\u00e1ciu odtoku.<\/p>\n<p>Tak teda, o obdobiach neaktivity.<\/p>\n<p>Na formulovanie defin\u00edcie odtoku je potrebn\u00e9 vybra\u0165 obdobie neaktivity. Aby sme vytvorili predpove\u010f odtoku v okamihu <img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/d2faf7d1a11c2770987ab6316b1db5e8.png\" style=\"display:block;margin: 0 auto;\" \/>, je potrebn\u00e9 ma\u0165 hist\u00f3riu klientov minim\u00e1lne na 3 mesiace v intervale <img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/194e5601c2f0f49866d8e8e5a9ff5993.png\" style=\"display:block;margin: 0 auto;\" \/>. Hist\u00f3ria bola obmedzen\u00e1 na 19 mesiacov, preto sme sa rozhodli zobra\u0165 obdobie neaktivity na 6 mesiacov, ak existuje. Ako minim\u00e1lne obdobie pre kvalitn\u00fa predpove\u010f sme zobrali 3 mesiace. \u010c\u00edsla 3 a 6 mesiacov sme vybrali emp\u00edricky na z\u00e1klade anal\u00fdzy spr\u00e1vania t\u00fdchto klientov.<\/p>\n<p>Defin\u00edciu odtoku sme formulovali takto: mesiac odtoku klienta <img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/3d581f5378975105aa6e73cacc175aeb.png\" style=\"display:block;margin: 0 auto;\" \/> je prv\u00fd mesiac s ACTIVE_FLAG=0, kde od tohto mesiaca nasleduje minim\u00e1lne \u0161es\u0165 n\u00fal za sebou v poli ACTIVE_FLAG, in\u00fdmi slovami, mesiac, po\u010d\u00ednaj\u00fac ktor\u00fdm klient bol 6 mesiacov neakt\u00edvny. <\/p>\n<p><img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/04ea6d4c59a9497733e4a181a4a5847c.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<i>Po\u010det od\u00edden\u00fdch klientov<\/i><\/p>\n<p><img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/4d470cbe91bbc6ee7ddfd1933e90e62c.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<i>Po\u010det zost\u00e1vaj\u00facich klientov<\/i><\/p>\n<h2>Ako sa predpoklad\u00e1, odtok<\/h2>\n<p>\nV podobn\u00fdch s\u00fa\u0165a\u017eiach, a vo v\u0161eobecnosti na praxi, sa odtok celkom \u010dasto predpoved\u00e1 t\u00fdmto sp\u00f4sobom. Klient vyu\u017e\u00edva produkty a slu\u017eby v r\u00f4znych \u010dasov\u00fdch intervaloch, \u00fadaje o interakcii s n\u00edm s\u00fa predstavovan\u00e9 vo forme vektora znakov pevn\u00fdch d\u013a\u017eok n. Naj\u010dastej\u0161ie sa do tejto inform\u00e1cie zahr\u0148uje:<\/p>\n<ul>\n<li>Charakterizuj\u00face \u00fadaje o pou\u017e\u00edvate\u013eovi (demografick\u00e9 \u00fadaje, marketingov\u00fd segment).<\/li>\n<li>Hist\u00f3ria pou\u017e\u00edvania bankov\u00fdch produktov a slu\u017eieb (to s\u00fa akcie klientov, ktor\u00e9 s\u00fa v\u017edy viazan\u00e9 na konkr\u00e9tny \u010das alebo po\u017eadovan\u00fa \u010dasov\u00fa peri\u00f3du).<\/li>\n<li>Extern\u00e9 d\u00e1ta, ak sa ich podarilo z\u00edska\u0165 \u2014 napr\u00edklad recenzie zo soci\u00e1lnych siet\u00ed.<\/li>\n<\/ul>\n<p>\nA a\u017e po tom sa vypo\u010d\u00edta defin\u00edcia odchodu, vlastn\u00e1 pre ka\u017ed\u00fa \u00falohu. Potom sa pou\u017e\u00edva algoritmus strojov\u00e9ho u\u010denia, ktor\u00fd predpoved\u00e1 pravdepodobnos\u0165 odchodu z\u00e1kazn\u00edka. <img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/79ce2d7f3193a9aab0b985c9a0cd0ee1.png\" style=\"display:block;margin: 0 auto;\" \/> na z\u00e1klade vektora faktorov. <img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/546791c513159e41d5950adca001d620.png\" style=\"display:block;margin: 0 auto;\" \/>Na tr\u00e9ning algoritmu sa vyu\u017e\u00edva jeden z najzn\u00e1mej\u0161\u00edch r\u00e1mcov na vytv\u00e1ranie ensemble rozhodovac\u00edch stromov, <noindex><a rel=\"nofollow\" href=\"https:\/\/xgboost.readthedocs.io\/en\/latest\/\">XGBoost<\/a><\/noindex>, <noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/Microsoft\/LightGBM\">LightGBM<\/a><\/noindex>, <noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/catboost\/catboost\">CatBoost<\/a><\/noindex> alebo ich modifik\u00e1cie. <\/p>\n<p>Samotn\u00fd algoritmus nie je zl\u00fd, ale v oblasti predpovedania odchodu m\u00e1 nieko\u013eko v\u00e1\u017enych nedostatkov.<\/p>\n<ul>\n<li><b>Nem\u00e1 takzvan\u00fa \u201epam\u00e4\u0165\u201c.<\/b>Na vstup modelu prich\u00e1dza stanoven\u00fd po\u010det znakov, ktor\u00e9 zodpovedaj\u00fa aktu\u00e1lnemu \u010dasov\u00e9mu okamihu. Aby bolo mo\u017en\u00e9 zahrn\u00fa\u0165 inform\u00e1cie o hist\u00f3rii zmien parametrov, je potrebn\u00e9 vypo\u010d\u00edtava\u0165 \u0161peci\u00e1lne znaky, ktor\u00e9 charakterizuj\u00fa zmeny parametrov v \u010dase, napr\u00edklad po\u010det alebo s\u00fa\u010det bankov\u00fdch transakci\u00ed za posledn\u00e9 1, 2, 3 mesiace. Tak\u00fdto pr\u00edstup m\u00f4\u017ee len \u010diasto\u010dne odr\u00e1\u017ea\u0165 charakter \u010dasov\u00fdch zmien. <\/li>\n<li><b>Fixovan\u00fd horizont predpovedania.<\/b> Model je schopn\u00fd predpoveda\u0165 odchod z\u00e1kazn\u00edkov iba na vopred stanoven\u00e9 obdobie, napr\u00edklad predpove\u010f na jeden mesiac dopredu. Ak je potrebn\u00e1 predpove\u010f na in\u00e9 obdobie, napr\u00edklad na tri mesiace, je potrebn\u00e9 znovu vytvori\u0165 tr\u00e9ningov\u00fa vzorku a pretr\u00e9nova\u0165 nov\u00fd model. <\/li>\n<\/ul>\n<p><\/p>\n<h2>N\u00e1\u0161 pr\u00edstup<\/h2>\n<p>\nRozhodli sme sa, \u017ee nebudeme pou\u017e\u00edva\u0165 \u0161tandardn\u00e9 pr\u00edstupy. V s\u00fa\u0165a\u017ei sa okrem n\u00e1s zaregistrovalo \u010fal\u0161\u00edch 497 \u013eud\u00ed, pri\u010dom ka\u017ed\u00fd z nich mal bohat\u00e9 sk\u00fasenosti. Tak\u017ee sna\u017ei\u0165 sa robi\u0165 nie\u010do pod\u013ea \u0161tandardn\u00fdch sch\u00e9m v tak\u00fdchto podmienkach \u2014 nie je najlep\u0161ia my\u0161lienka.<\/p>\n<p>A za\u010dali sme rie\u0161i\u0165 probl\u00e9my, ktor\u00e9 stoja pred modelom bin\u00e1rnej klasifik\u00e1cie, pomocou predpovedania pravdepodobnostn\u00e9ho rozdelenia \u010dasu odchodu z\u00e1kazn\u00edkov. Tak\u00fdto pr\u00edstup si m\u00f4\u017eete pozrie\u0165, <noindex><a rel=\"nofollow\" href=\"https:\/\/ragulpr.github.io\/assets\/draft_master_thesis_martinsson_egil_wtte_rnn_2016.pdf\">tu<\/a><\/noindex>umo\u017e\u0148uje flexibilnej\u0161ie predpoveda\u0165 odchod a overova\u0165 zlo\u017eitej\u0161ie hypot\u00e9zy ako v klasickom pr\u00edstupe. Ako rodinu rozdelen\u00ed modeluj\u00facich \u010das odchodu sme si vybrali rozdelenie <noindex><a rel=\"nofollow\" href=\"https:\/\/ru.wikipedia.org\/wiki\/%D0%A0%D0%B0%D1%81%D0%BF%D1%80%D0%B5%D0%B4%D0%B5%D0%BB%D0%B5%D0%BD%D0%B8%D0%B5_%D0%92%D0%B5%D0%B9%D0%B1%D1%83%D0%BB%D0%BB%D0%B0\">Weibull<\/a><\/noindex> pre jeho \u0161irok\u00e9 uplatnenie v anal\u00fdze pre\u017eitia. Spr\u00e1vanie z\u00e1kazn\u00edka sa d\u00e1 pova\u017eova\u0165 za ak\u00fdsi druh pre\u017eitia. <\/p>\n<p>Tu s\u00fa pr\u00edklady rozdelenia pravdepodobnostn\u00fdch hust\u00f4t Weibull v z\u00e1vislosti od parametrov. <img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/57c935da52faf825ceedb4389d21c354.png\" style=\"display:block;margin: 0 auto;\" \/> a <img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/85d1c92d5b9dfa83d13092edefb29a09.png\" style=\"display:block;margin: 0 auto;\" \/>:<\/p>\n<p><img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/371e51ab8df2c10d5b139aadf290ebf8.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<br \/>\nToto je hustota rozdelenia pravdepodobnosti odtoku klienta troch r\u00f4znych klientov v priebehu \u010dasu. \u010cas je zobrazen\u00fd v mesiacoch. In\u00fdmi slovami, tento graf ukazuje, kedy je najpravdepodobnej\u0161ie, \u017ee d\u00f4jde k odtoku klienta v nasleduj\u00facich dvoch mesiacoch. Ako je vidie\u0165, klient s rozdelen\u00edm m\u00e1 v\u00e4\u010d\u0161\u00ed potenci\u00e1l od\u00eds\u0165 sk\u00f4r ako klienti s rozdeleniami Weibull(2, 0.5) a Weibull(3,1). <\/p>\n<p>Nakoniec sa vytvor\u00ed model, ktor\u00fd predpoved\u00e1 pre ka\u017ed\u00e9ho klienta pre ak\u00fdko\u013evek<br \/>\nmesiac parametre rozdelenia Weibull, ktor\u00e9 najlep\u0161ie odr\u00e1\u017eaj\u00fa pravdepodobnos\u0165 odtoku v priebehu \u010dasu. Ak hovor\u00edme podrobnej\u0161ie: <\/p>\n<ul>\n<li>Cie\u013eov\u00e9 znaky na tr\u00e9novacej vzorke \u2014 \u010das, ktor\u00fd zost\u00e1va do odtoku v konkr\u00e9tnom mesiaci pre konkr\u00e9tneho klienta.<\/li>\n<li>Ak ukazovate\u013e odtoku pre klienta nie je k dispoz\u00edcii, predpoklad\u00e1me, \u017ee \u010das odtoku je dlh\u0161\u00ed ako po\u010det mesiacov od aktu\u00e1lneho \u010dasu do konca dostupnej hist\u00f3rie.<\/li>\n<li>Pou\u017eit\u00fd model: rekurentn\u00e1 neur\u00f3nov\u00e1 sie\u0165 s vrstvou LSTM.<\/li>\n<li>Ako funkciu straty pou\u017e\u00edvame negat\u00edvnu logaritmick\u00fa funkciu pravdepodobnosti pre rozdelenie Weibull.<\/li>\n<\/ul>\n<p>\nTu s\u00fa v\u00fdhody tohto pr\u00edstupu: <\/p>\n<ul>\n<li>Pravdepodobnostn\u00e9 rozdelenie, okrem zjavnej mo\u017enosti bin\u00e1rnej klasifik\u00e1cie, umo\u017e\u0148uje flexibilne predpoveda\u0165 r\u00f4zne udalosti, napr\u00edklad, \u010di prestane klient vyu\u017e\u00edva\u0165 slu\u017eby banky v priebehu 3 mesiacov. Taktie\u017e, ak je to potrebn\u00e9, mo\u017eno pri tomto rozdelen\u00ed priemerova\u0165 r\u00f4zne metriky.<\/li>\n<li>Rekurentn\u00e1 neur\u00f3nov\u00e1 sie\u0165 LSTM m\u00e1 pam\u00e4\u0165 a efekt\u00edvne vyu\u017e\u00edva cel\u00fa dostupn\u00fa hist\u00f3riu. S roz\u0161\u00edren\u00edm alebo objasnen\u00edm hist\u00f3rie presnos\u0165 rastie.<\/li>\n<li>Pr\u00edstup je mo\u017en\u00e9 bez probl\u00e9mov \u0161k\u00e1lova\u0165 pri rozdelen\u00ed \u010dasov\u00fdch intervalov na men\u0161ie (napr\u00edklad pri rozdelen\u00ed mesiacov na t\u00fd\u017edne).<\/li>\n<\/ul>\n<p>\nAle nesta\u010d\u00ed len vytvori\u0165 dobr\u00fd model, je potrebn\u00e9 aj spr\u00e1vne pos\u00fadi\u0165 jeho kvalitu.<\/p>\n<h2>Ako sme hodnotili kvalitu<\/h2>\n<p>\nAko metriku sme zvolili Lift Curve. Pou\u017e\u00edva sa v podnikan\u00ed na podobn\u00e9 pr\u00edpady kv\u00f4li zrozumite\u013enej interpret\u00e1cii, je dobre pop\u00edsan\u00e1 <noindex><a rel=\"nofollow\" href=\"http:\/\/www2.cs.uregina.ca\/~dbd\/cs831\/notes\/lift_chart\/lift_chart.html\">tu<\/a><\/noindex> a <noindex><a rel=\"nofollow\" href=\"https:\/\/www.information-management.com\/news\/why-lift\">tu<\/a><\/noindex>. Ak by sme jednou vetou op\u00edsali v\u00fdznam tejto metriky, dostaneme: \"O ko\u013eko viac algoritmus rob\u00ed lep\u0161iu predpove\u010f na prv\u00fdch <img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/72818f84f73c2838fe7d63b33a96fd7b.png\" style=\"display:block;margin: 0 auto;\" \/>% ako n\u00e1hodne.\" <\/p>\n<h2>Tr\u00e9nujeme modely<\/h2>\n<p>\nPodmienkami s\u00fa\u0165a\u017ee nebola stanoven\u00e1 konkr\u00e9tna metrika kvality, pod\u013ea ktorej by bolo mo\u017en\u00e9 porovn\u00e1va\u0165 r\u00f4zne modely a pr\u00edstupy. Navy\u0161e, defin\u00edcia pojmu odtoku m\u00f4\u017ee by\u0165 r\u00f4zna a m\u00f4\u017ee z\u00e1visie\u0165 od zadania \u00falohy, ktor\u00e1 je op\u00e4\u0165 definovan\u00e1 obchodn\u00fdmi cie\u013emi. Preto, aby sme zistili, ktor\u00fd pr\u00edstup je lep\u0161\u00ed, vytr\u00e9novali sme dva modely:<\/p>\n<ol>\n<li>\u010casto pou\u017e\u00edvan\u00fd pr\u00edstup bin\u00e1rnej klasifik\u00e1cie s pou\u017eit\u00edm algoritmu strojov\u00e9ho u\u010denia zoskupenia rozhodovac\u00edch stromov (<noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/Microsoft\/LightGBM\">LightGBM<\/a><\/noindex>);<\/li>\n<li> Model Weibull-LSTM<\/li>\n<\/ol>\n<p>\nTestovacia vzorka pozost\u00e1vala z 500 vopred vybran\u00fdch klientov, ktor\u00ed neboli v tr\u00e9ningovej vzorke. Pre model sa vykonalo ladenie hyperparameterov s pou\u017eit\u00edm kr\u00ed\u017eovej valid\u00e1cie rozdelenej pod\u013ea klientov. Pre tr\u00e9ning ka\u017ed\u00e9ho modelu sa pou\u017e\u00edvali rovnak\u00e9 s\u00fabory znakov. <\/p>\n<p>V s\u00favislosti s t\u00fdm, \u017ee model nem\u00e1 pam\u00e4\u0165, sa pre neho zoh\u013eadnili \u0161peci\u00e1lne znaky, ktor\u00e9 ukazovali pomer zmeny parametrov jedn\u00e9ho mesiaca k priemernej hodnote parametrov za posledn\u00e9 tri mesiace. Toto charakterizovalo r\u00fdchlos\u0165 zmeny hodn\u00f4t za posledn\u00e9 trojmesa\u010dn\u00e9 obdobie. Bez toho by bol model zalo\u017een\u00fd na Random Forest vopred v nev\u00fdhodnej poz\u00edcii v porovnan\u00ed s Weibull-LSTM.<\/p>\n<h2>Pre\u010do je LSTM s Weibullov\u00fdm rozdelen\u00edm lep\u0161ie ako pr\u00edstup zalo\u017een\u00fd na zoskupen\u00ed rozhodovac\u00edch stromov <\/h2>\n<p>\nTu je v\u0161etko jasn\u00e9 doslova p\u00e1r obr\u00e1zkami.<\/p>\n<p><img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/0a87a27804f1b285f66a889ed798cf6b.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<i>Porovnanie Lift Curve pre klasick\u00fd algoritmus a Weibull-LSTM<\/i><\/p>\n<p><img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/6c281677260098c7993a690c923692c8.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<i>Porovnanie metriky Lift Curve pod\u013ea mesiacov pre klasick\u00fd algoritmus a Weibull-LSTM<\/i><\/p>\n<p>Vo v\u0161eobecnosti LSTM prekon\u00e1va klasick\u00fd algoritmus takmer vo v\u0161etk\u00fdch pr\u00edpadoch.<\/p>\n<h2>Predpokladanie odtoku<\/h2>\n<p>\nModel zalo\u017een\u00fd na rekurentnej neur\u00f3novej sieti s LSTM bunkami s Weibullov\u00fdm rozdelen\u00edm m\u00f4\u017ee vopred predpoveda\u0165 odtok, napr\u00edklad predpoveda\u0165 odchod klienta v nasleduj\u00facich n mesiacoch. Zv\u00e1\u017eme pr\u00edpad pre n = 3. V tomto pr\u00edpade by neur\u00f3nov\u00e1 sie\u0165 mala spr\u00e1vne ur\u010di\u0165: \u010di klient od\u00edde od nasleduj\u00faceho mesiaca a\u017e po n. mesiac. In\u00fdmi slovami, mala by spr\u00e1vne ur\u010di\u0165, \u010di klient zostane po uplynut\u00ed n mesiacov. To mo\u017eno pova\u017eova\u0165 za predpove\u010f vopred: predpove\u010f okamihu, ke\u010f klient za\u010dal uva\u017eova\u0165 o odchode. <\/p>\n<p>Porovnajme Lift Curve pre Weibull-LSTM na 1, 2 a 3 mesiace pred odtokom:<\/p>\n<p><img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/4b2ed9cb21844cf5dc8f54bcfe092a9d.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<br \/>\nAko sme u\u017e spom\u00ednali vy\u0161\u0161ie, d\u00f4le\u017eit\u00e9 s\u00fa aj predpovede pre klientov, ktor\u00ed na nejak\u00fd \u010das vypadn\u00fa z akt\u00edvneho pou\u017e\u00edvania. Preto do n\u00e1\u0161ho v\u00fdberu prid\u00e1me aj pr\u00edpady, ke\u010f klient, ktor\u00fd odi\u0161iel, bol neakt\u00edvny jeden alebo dva mesiace a over\u00edme, \u010di Weibull-LSTM spr\u00e1vne klasifikuje tak\u00e9to pr\u00edpady ako odchod. Ke\u010f\u017ee tak\u00e9to pr\u00edpady sa u\u017e vyskytli vo v\u00fdbere, o\u010dak\u00e1vame, \u017ee sie\u0165 s nimi dobre zaobch\u00e1dza:<\/p>\n<p><img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/4fe6e918aff2f16758c671c8d547895d.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<\/p>\n<h2>Udr\u017eanie klientov<\/h2>\n<p>\nVlastne, to je hlavn\u00e1 vec, ktor\u00fa m\u00f4\u017eeme urobi\u0165, ke\u010f m\u00e1me inform\u00e1cie o tom, \u017ee niektor\u00ed klienti sa chystaj\u00fa presta\u0165 pou\u017e\u00edva\u0165 produkt. Pri vytv\u00e1ran\u00ed modelu, ktor\u00fd by mohol pon\u00faknu\u0165 nie\u010do u\u017eito\u010dn\u00e9 klientom s cie\u013eom ich udr\u017ea\u0165 - to nebude mo\u017en\u00e9 urobi\u0165, ak m\u00e1te len ve\u013emi m\u00e1lo sk\u00fasenost\u00ed s podobn\u00fdmi pokusmi, ktor\u00e9 skon\u010dili \u00faspe\u0161ne.<\/p>\n<p>Nemali sme tak\u00fato hist\u00f3riu, a preto sme to rozhodli takto.<\/p>\n<ol>\n<li>Vytv\u00e1rame model, ktor\u00fd ur\u010duje zauj\u00edmav\u00e9 produkty pre ka\u017ed\u00e9ho klienta.<\/li>\n<li>Ka\u017ed\u00fd mesiac sp\u00fa\u0161\u0165ame klasifik\u00e1tor a ur\u010dujeme potenci\u00e1lne odch\u00e1dzaj\u00facich klientov.<\/li>\n<li>\u010casti klientov pon\u00fakame produkt pod\u013ea modelu z bodu 1 a zaznamen\u00e1vame svoje akcie.<\/li>\n<li>Po nieko\u013ek\u00fdch mesiacoch sledujeme, ktor\u00ed z t\u00fdchto potenci\u00e1lne odch\u00e1dzaj\u00facich klientov odi\u0161li a ktor\u00ed zostali. T\u00fdmto sp\u00f4sobom formujeme tr\u00e9ningov\u00fd s\u00fabor.<\/li>\n<li>Tr\u00e9nujeme model na hist\u00f3rii z\u00edskanej v bode 4.<\/li>\n<li>Volite\u013ene, opakujeme postup, vymie\u0148aj\u00fac model z bodu 1 za model z\u00edskan\u00fd v bode 5.<\/li>\n<\/ol>\n<p>\nKontrolou kvality tak\u00e9hoto udr\u017eania m\u00f4\u017ee by\u0165 be\u017en\u00e9 A\/B testovanie - rozdel\u00edme klientov, ktor\u00ed s\u00fa potenci\u00e1lne odch\u00e1dzaj\u00faci, do dvoch skup\u00edn. Jednej pon\u00fakame produkty na z\u00e1klade n\u00e1\u0161ho modelu udr\u017eania, druhej nenab\u00edzame ni\u010d. Rozhodli sme sa tr\u00e9nova\u0165 model, ktor\u00fd by mohol prinies\u0165 \u00fa\u017eitok u\u017e v bode 1 n\u00e1\u0161ho pr\u00edkladu.<\/p>\n<p>Chceli sme, aby bola segment\u00e1cia \u010do najviac interpretovate\u013en\u00e1. Na tento \u00fa\u010del sme vybrali nieko\u013eko znakov, ktor\u00e9 mohli by\u0165 \u013eahko interpretovate\u013en\u00e9: celkov\u00fd po\u010det transakci\u00ed, plat, celkov\u00fd obrat na \u00fa\u010dte, vek, pohlavie. Znaky z tabu\u013eky \u201eKarty\u201c sme nezoh\u013eadnili ako m\u00e1lo informat\u00edvne, a znaky z tabu\u013eky 3 \u201eZmluvy\u201c - z d\u00f4vodu zlo\u017eitosti spracovania, aby sa predi\u0161lo \u00faniku \u00fadajov medzi valida\u010dn\u00fdm a tr\u00e9ningov\u00fdm s\u00faborom.<\/p>\n<p>Klastr\u00e1cia bola vykonan\u00e1 pomocou modelov Gaussian mixture. Informa\u010dn\u00e9 krit\u00e9rium Akaike umo\u017enilo ur\u010di\u0165 2 optim\u00e1. Prv\u00e9 optimum zodpoved\u00e1 1 klustru. Druh\u00e9 optimum, menej vyjadren\u00e9, zodpoved\u00e1 80 klastrom. Z t\u00fdchto v\u00fdsledkov m\u00f4\u017eeme vyvodi\u0165 nasleduj\u00faci z\u00e1ver: je ve\u013emi \u0165a\u017ek\u00e9 rozdeli\u0165 d\u00e1ta do klastrov bez a priori zadan\u00fdch inform\u00e1ci\u00ed. Na kvalitnej\u0161iu klasifik\u00e1ciu s\u00fa potrebn\u00e9 d\u00e1ta, ktor\u00e9 podrobne opisuj\u00fa ka\u017ed\u00e9ho klienta.<\/p>\n<p>Preto bola zv\u00e1\u017een\u00e1 \u00faloha u\u010denia pod doh\u013eadom, aby sme ka\u017ed\u00e9mu jednotliv\u00e9mu klientovi pon\u00fakli jeho vlastn\u00fd produkt. Zva\u017eovali sa nasleduj\u00face produkty: \u201eTerm\u00ednovan\u00fd vklad\u201c, \u201eKreditn\u00e1 karta\u201c, \u201eOvdraf\u201c, \u201eSpotrebn\u00fd \u00faver\u201c, \u201eAuto\u00faver\u201c, \u201eHypot\u00e9ka\u201c. <\/p>\n<p>V d\u00e1tach bol pr\u00edtomn\u00fd e\u0161te jeden typ produktu: \u201eBe\u017en\u00fd \u00fa\u010det\u201c. Ale ten sme nezva\u017eovali kv\u00f4li nedostato\u010dnej informat\u00edvnosti. Na z\u00e1kladne pou\u017e\u00edvate\u013eov, ktor\u00ed s\u00fa klientami banky, t.j. nezru\u0161ili pou\u017e\u00edvanie jej produktov, bola postaven\u00e1 model, ktor\u00fd predpoved\u00e1, ak\u00fd produkt by ich mohol zauj\u00edma\u0165. Ako model bola zvolen\u00e1 logistick\u00e1 regresia a ako metrika hodnotenia kvality sa pou\u017e\u00edvala hodnota Lift pre prv\u00fdch 10 percentilov. <\/p>\n<p>Kvalitu modelu je mo\u017en\u00e9 oceni\u0165 na obr\u00e1zku.<\/p>\n<p><img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/77cab18988ab9c3b768b86e31479817f.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<i>V\u00fdsledky modelu odpor\u00fa\u010dania produktov pre klientov <\/i> <\/p>\n<h2>Z\u00e1ver<\/h2>\n<p>\nTento pr\u00edstup n\u00e1m priniesol prv\u00e9 miesto v kateg\u00f3rii \u201eAI v bank\u00e1ch\u201c na \u0160ampion\u00e1te AI RAIF-Challenge 2017.<\/p>\n<p><img decoding=\"async\" alt=\"Ako sme predpovedali odliv, pristupuj\u00fac k nemu ako k pr\u00edrodnej katastrofe\" src=\"\/wp-content\/uploads\/2019\/03\/0c321869dfd02323766c025ed4021bcf.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<br \/>\nZjavne bolo najd\u00f4le\u017eitej\u0161ie prist\u00fapi\u0165 k probl\u00e9mu z menej tradi\u010dnej str\u00e1nky a pou\u017ei\u0165 met\u00f3du, ktor\u00e1 sa zvy\u010dajne pou\u017e\u00edva v in\u00fdch situ\u00e1ci\u00e1ch.<\/p>\n<p>Aj ke\u010f hromadn\u00fd odklon pou\u017e\u00edvate\u013eov m\u00f4\u017ee by\u0165 pre slu\u017eby \u017eivelnou pohromou.<\/p>\n<p>T\u00fato met\u00f3du je mo\u017en\u00e9 vzia\u0165 na vedomie aj pre ak\u00fako\u013evek in\u00fa oblas\u0165, kde je d\u00f4le\u017eit\u00e9 zoh\u013eadni\u0165 odliv, nielen v bank\u00e1ch. Napr\u00edklad sme ju pou\u017eili aj na vypo\u010d\u00edtanie vlastn\u00e9ho odlivu - v Sibirskom a Petersburgskom pobo\u010dke Rostelekom.<\/p>\n<p><i>\u201eLaborat\u00f3rium inteligentnej anal\u00fdzy d\u00e1t\u201c spolo\u010dnos\u0165 \u201eVyh\u013ead\u00e1vac\u00ed port\u00e1l Sputnik\u201c<\/i><br \/>\n<br \/>Zdroj: <a content=\"nofollow\" rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/company\/rostelecom\/blog\/444694\/\">habr.com<\/a><\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"<p>\u0418\u043d\u043e\u0433\u0434\u0430 \u0434\u043b\u044f \u0442\u043e\u0433\u043e, \u0447\u0442\u043e\u0431\u044b \u0440\u0435\u0448\u0438\u0442\u044c \u043a\u0430\u043a\u0443\u044e-\u0442\u043e \u043f\u0440\u043e\u0431\u043b\u0435\u043c\u0443, \u043d\u0430\u0434\u043e \u043f\u0440\u043e\u0441\u0442\u043e \u0432\u0437\u0433\u043b\u044f\u043d\u0443\u0442\u044c \u043d\u0430 \u043d\u0435\u0435 \u043f\u043e\u0434 \u0434\u0440\u0443\u0433\u0438\u043c \u0443\u0433\u043b\u043e\u043c. \u0414\u0430\u0436\u0435 \u0435\u0441\u043b\u0438 \u043f\u043e\u0441\u043b\u0435\u0434\u043d\u0438\u0435 \u043b\u0435\u0442 10 \u043f\u043e\u0434\u043e\u0431\u043d\u044b\u0435 \u043f\u0440\u043e\u0431\u043b\u0435\u043c\u044b \u0440\u0435\u0448\u0430\u043b\u0438 \u043e\u0434\u043d\u0438\u043c \u0438 \u0442\u0435\u043c \u0436\u0435 \u0441\u043f\u043e\u0441\u043e\u0431\u043e\u043c \u0441 \u0440\u0430\u0437\u043d\u044b\u043c \u044d\u0444\u0444\u0435\u043a\u0442\u043e\u043c, \u043d\u0435 \u0444\u0430\u043a\u0442, \u0447\u0442\u043e \u044d\u0442\u043e\u0442 \u0441\u043f\u043e\u0441\u043e\u0431 \u0435\u0434\u0438\u043d\u0441\u0442\u0432\u0435\u043d\u043d\u044b\u0439. \u0415\u0441\u0442\u044c \u0442\u0430\u043a\u0430\u044f \u0442\u0435\u043c\u0430, \u043a\u0430\u043a \u043e\u0442\u0442\u043e\u043a \u043a\u043b\u0438\u0435\u043d\u0442\u043e\u0432. \u0428\u0442\u0443\u043a\u0430 \u043d\u0435\u0438\u0437\u0431\u0435\u0436\u043d\u0430\u044f, \u043f\u043e\u0442\u043e\u043c\u0443 \u0447\u0442\u043e \u043a\u043b\u0438\u0435\u043d\u0442\u044b \u043b\u044e\u0431\u043e\u0439 \u043a\u043e\u043c\u043f\u0430\u043d\u0438\u0438 \u043c\u043e\u0433\u0443\u0442 \u043f\u043e \u043c\u043d\u043e\u0436\u0435\u0441\u0442\u0432\u0443 \u043f\u0440\u0438\u0447\u0438\u043d \u0432\u0437\u044f\u0442\u044c [&hellip;]<\/p>\n","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[],"tags":[],"class_list":["post-30044","post","type-post","status-publish","format-standard","hentry"],"aioseo_notices":[],"aioseo_head":"\n\t\t<!-- All in One SEO 5.0.4 - aioseo.com -->\n\t<meta name=\"description\" content=\"\u0418\u043d\u043e\u0433\u0434\u0430 \u0434\u043b\u044f \u0442\u043e\u0433\u043e, \u0447\u0442\u043e\u0431\u044b \u0440\u0435\u0448\u0438\u0442\u044c \u043a\u0430\u043a\u0443\u044e-\u0442\u043e \u043f\u0440\u043e\u0431\u043b\u0435\u043c\u0443, \u043d\u0430\u0434\u043e \u043f\u0440\u043e\u0441\u0442\u043e \u0432\u0437\u0433\u043b\u044f\u043d\u0443\u0442\u044c \u043d\u0430 \u043d\u0435\u0435 \u043f\u043e\u0434 \u0434\u0440\u0443\u0433\u0438\u043c \u0443\u0433\u043b\u043e\u043c.\" \/>\n\t<meta name=\"robots\" content=\"max-image-preview:large\" \/>\n\t<meta name=\"author\" content=\"Yuri Gagarin\"\/>\n\t<link rel=\"canonical\" href=\"https:\/\/prohoster.info\/sk\/blog\/kak-my-predskazyvali-ottok-podojdya-k-nemu-kak-k-stihijnomu-bedstviyu\" \/>\n\t\t<meta name=\"generator\" content=\"All in One SEO (AIOSEO) 5.0.4\" \/>\n\t\t<meta property=\"og:locale\" content=\"sk_SK\" \/>\n\t\t<meta property=\"og:site_name\" content=\"ProHoster | \u041a\u0443\u043f\u0438\u0442\u044c \u043d\u0430\u0434\u0435\u0436\u043d\u044b\u0439 \u0445\u043e\u0441\u0442\u0438\u043d\u0433 \u0434\u043b\u044f \u0441\u0430\u0439\u0442\u043e\u0432 \u0441 \u0437\u0430\u0449\u0438\u0442\u043e\u0439 \u043e\u0442 DDoS, VPS VDS \u0441\u0435\u0440\u0432\u0435\u0440\u044b\" \/>\n\t\t<meta property=\"og:type\" content=\"article\" \/>\n\t\t<meta property=\"og:title\" content=\"\ud83e\udd47\u041a\u0430\u043a \u043c\u044b \u043f\u0440\u0435\u0434\u0441\u043a\u0430\u0437\u044b\u0432\u0430\u043b\u0438 \u043e\u0442\u0442\u043e\u043a, \u043f\u043e\u0434\u043e\u0439\u0434\u044f \u043a \u043d\u0435\u043c\u0443 \u043a\u0430\u043a \u043a \u0441\u0442\u0438\u0445\u0438\u0439\u043d\u043e\u043c\u0443 \u0431\u0435\u0434\u0441\u0442\u0432\u0438\u044e | ProHoster\" \/>\n\t\t<meta property=\"og:description\" content=\"\u0418\u043d\u043e\u0433\u0434\u0430 \u0434\u043b\u044f \u0442\u043e\u0433\u043e, \u0447\u0442\u043e\u0431\u044b \u0440\u0435\u0448\u0438\u0442\u044c \u043a\u0430\u043a\u0443\u044e-\u0442\u043e \u043f\u0440\u043e\u0431\u043b\u0435\u043c\u0443, \u043d\u0430\u0434\u043e \u043f\u0440\u043e\u0441\u0442\u043e \u0432\u0437\u0433\u043b\u044f\u043d\u0443\u0442\u044c \u043d\u0430 \u043d\u0435\u0435 \u043f\u043e\u0434 \u0434\u0440\u0443\u0433\u0438\u043c \u0443\u0433\u043b\u043e\u043c.\" \/>\n\t\t<meta property=\"og:url\" content=\"https:\/\/prohoster.info\/sk\/blog\/kak-my-predskazyvali-ottok-podojdya-k-nemu-kak-k-stihijnomu-bedstviyu\" \/>\n\t\t<meta property=\"og:image\" content=\"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg\" \/>\n\t\t<meta property=\"og:image:secure_url\" content=\"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg\" \/>\n\t\t<meta property=\"og:image:width\" content=\"350\" \/>\n\t\t<meta property=\"og:image:height\" content=\"350\" \/>\n\t\t<meta property=\"article:published_time\" content=\"2019-10-31T18:33:22+00:00\" \/>\n\t\t<meta property=\"article:modified_time\" content=\"2019-10-31T18:33:22+00:00\" \/>\n\t\t<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/prohoster\" \/>\n\t\t<meta property=\"article:author\" content=\"https:\/\/www.facebook.com\/prohoster\" \/>\n\t\t<!-- All in One SEO -->\n\n","aioseo_head_json":{"title":"\ud83e\udd47Ako sme predpovedali odliv, prist\u00fapili k nemu ako k \u017eiveln\u00e9mu ne\u0161\u0165astiu | ProHoster","description":"Niekedy na vyrie\u0161enie probl\u00e9mu sta\u010d\u00ed pozrie\u0165 sa na neho z in\u00e9ho uhla.","canonical_url":"https:\/\/prohoster.info\/sk\/blog\/kak-my-predskazyvali-ottok-podojdya-k-nemu-kak-k-stihijnomu-bedstviyu","robots":"max-image-preview:large","keywords":"","webmasterTools":{"miscellaneous":""},"schema":null,"og:locale":"sk_SK","og:site_name":"ProHoster | \u041a\u0443\u043f\u0438\u0442\u044c \u043d\u0430\u0434\u0435\u0436\u043d\u044b\u0439 \u0445\u043e\u0441\u0442\u0438\u043d\u0433 \u0434\u043b\u044f \u0441\u0430\u0439\u0442\u043e\u0432 \u0441 \u0437\u0430\u0449\u0438\u0442\u043e\u0439 \u043e\u0442 DDoS, VPS VDS \u0441\u0435\u0440\u0432\u0435\u0440\u044b","og:type":"article","og:title":"\ud83e\udd47\u041a\u0430\u043a \u043c\u044b \u043f\u0440\u0435\u0434\u0441\u043a\u0430\u0437\u044b\u0432\u0430\u043b\u0438 \u043e\u0442\u0442\u043e\u043a, \u043f\u043e\u0434\u043e\u0439\u0434\u044f \u043a \u043d\u0435\u043c\u0443 \u043a\u0430\u043a \u043a \u0441\u0442\u0438\u0445\u0438\u0439\u043d\u043e\u043c\u0443 \u0431\u0435\u0434\u0441\u0442\u0432\u0438\u044e | ProHoster","og:description":"\u0418\u043d\u043e\u0433\u0434\u0430 \u0434\u043b\u044f \u0442\u043e\u0433\u043e, \u0447\u0442\u043e\u0431\u044b \u0440\u0435\u0448\u0438\u0442\u044c \u043a\u0430\u043a\u0443\u044e-\u0442\u043e \u043f\u0440\u043e\u0431\u043b\u0435\u043c\u0443, \u043d\u0430\u0434\u043e \u043f\u0440\u043e\u0441\u0442\u043e \u0432\u0437\u0433\u043b\u044f\u043d\u0443\u0442\u044c \u043d\u0430 \u043d\u0435\u0435 \u043f\u043e\u0434 \u0434\u0440\u0443\u0433\u0438\u043c \u0443\u0433\u043b\u043e\u043c.","og:url":"https:\/\/prohoster.info\/sk\/blog\/kak-my-predskazyvali-ottok-podojdya-k-nemu-kak-k-stihijnomu-bedstviyu","og:image":"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg","og:image:secure_url":"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg","og:image:width":350,"og:image:height":350,"article:published_time":"2019-10-31T18:33:22+00:00","article:modified_time":"2019-10-31T18:33:22+00:00","article:publisher":"https:\/\/www.facebook.com\/prohoster","article:author":"https:\/\/www.facebook.com\/prohoster"},"aioseo_meta_data":{"post_id":"30044","title":null,"description":null,"keywords":null,"keyphrases":null,"primary_term":null,"canonical_url":null,"og_title":null,"og_description":null,"og_object_type":"default","og_image_type":"default","og_image_url":null,"og_image_width":null,"og_image_height":null,"og_image_custom_url":null,"og_image_custom_fields":null,"og_video":null,"og_custom_url":null,"og_article_section":null,"og_article_tags":null,"twitter_use_og":false,"twitter_card":"default","twitter_image_type":"default","twitter_image_url":null,"twitter_image_custom_url":null,"twitter_image_custom_fields":null,"twitter_title":null,"twitter_description":null,"schema":{"blockGraphs":[],"customGraphs":[],"default":{"data":{"Article":[],"Course":[],"Dataset":[],"FAQPage":[],"Movie":[],"Person":[],"Product":[],"ProductReview":[],"Car":[],"Recipe":[],"Service":[],"SoftwareApplication":[],"WebPage":[]},"graphName":"Article","isEnabled":true},"graphs":[]},"schema_type":null,"schema_type_options":null,"pillar_content":false,"robots_default":true,"robots_noindex":false,"robots_noarchive":false,"robots_nosnippet":false,"robots_nofollow":false,"robots_noimageindex":false,"robots_noodp":false,"robots_notranslate":false,"robots_max_snippet":null,"robots_max_videopreview":null,"robots_max_imagepreview":"large","priority":null,"frequency":null,"local_seo":null,"seo_analyzer_scan_date":"2026-01-20 23:33:19","breadcrumb_settings":null,"limit_modified_date":false,"reviewed_by":null,"ai":null,"created":"2021-03-01 03:43:15","updated":"2026-01-20 23:33:19","focus_keyword":null,"additional_keywords":null,"truseo_locale":null},"gt_translate_keys":[{"key":"link","format":"url"}],"_links":{"self":[{"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/posts\/30044","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/comments?post=30044"}],"version-history":[{"count":0,"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/posts\/30044\/revisions"}],"wp:attachment":[{"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/media?parent=30044"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/categories?post=30044"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/tags?post=30044"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}