{"id":5628,"date":"2025-12-04T08:32:26","date_gmt":"2025-12-04T05:32:26","guid":{"rendered":"https:\/\/datakapital.com\/blog\/?p=5628"},"modified":"2025-12-04T08:32:26","modified_gmt":"2025-12-04T05:32:26","slug":"kredi-risk-yonetiminde-yapay-zeka-ve-makine-ogrenimi-uygulamalari","status":"publish","type":"post","link":"https:\/\/datakapital.com\/blog\/kredi-risk-yonetiminde-yapay-zeka-ve-makine-ogrenimi-uygulamalari\/","title":{"rendered":"Kredi Risk Y\u00f6netiminde Yapay Zeka ve Makine \u00d6\u011frenimi Uygulamalar\u0131"},"content":{"rendered":"<p><strong>Kredi risk y\u00f6netiminde yapay zeka ve makine \u00f6\u011frenimi\u00a0<\/strong>uygulamalar\u0131, bor\u00e7lananlar\u0131n risk durumlar\u0131n\u0131 \u00f6l\u00e7mek i\u00e7in kullan\u0131lan yeni jenerasyon finansal ara\u00e7lard\u0131r.\u00a0Kredi riski, bor\u00e7lunun borcunu \u00f6deyememe olas\u0131l\u0131\u011f\u0131 olarak tan\u0131mlan\u0131r ve riskin y\u00f6netimi, finansal sistemin temel dinami\u011fini olu\u015fturan kaynak fazlas\u0131 olan birimlerden kaynak a\u00e7\u0131\u011f\u0131 olan birimlere fon ak\u0131\u015flar\u0131n\u0131n etkinli\u011fi a\u00e7\u0131s\u0131ndan hem bireysel (t\u00fcketici) hem de kurumsal (ticari) kredilerde ileri d\u00fczey uygulamalar gerektirir. Bu uygulama aray\u0131\u015f\u0131, kredi riskinin -bor\u00e7lunun y\u00fck\u00fcml\u00fcl\u00fcklerini yerine getirememe ihtimalinin- do\u011fru \u00f6l\u00e7\u00fclmesi ihtiyac\u0131ndan gelir. Tarihsel s\u00fcre\u00e7te ise bu riskin y\u00f6netimi, kabaca bankac\u0131n\u0131n ki\u015fisel yarg\u0131s\u0131ndan istatistiksel tablolara, oradan da g\u00fcn\u00fcm\u00fcz\u00fcn karma\u015f\u0131k algoritmalar\u0131na do\u011fru evrilmi\u015ftir.<\/p>\n<p><strong>\u00a0<\/strong><strong>Geleneksel Yakla\u015f\u0131m ve S\u0131n\u0131rlar\u0131: &#8220;Kredinin 5C&#8217;si&#8221;nden Lojistik Regresyona<\/strong><\/p>\n<p>Geleneksel bankac\u0131l\u0131k, <a href=\"https:\/\/datakapital.com\/raporlar\/bist-te-islem-goren-mevduat-bankalarinin-camels-bilesenleri-uzerinden-degerlendirilmesi\">kredi de\u011ferlendirmesinde<\/a> uzun y\u0131llar boyunca <a href=\"https:\/\/www.findeks.com\/findeksakademi\/kredi-riski-nedir\" target=\"_blank\" rel=\"noopener\">&#8220;Kredinin 5C&#8217;si&#8221;<\/a> (Character, Capacity, Capital, Collateral, Conditions) olarak adland\u0131r\u0131lan prensibi dayanm\u0131\u015ft\u0131r. Bu niteliksel yakla\u015f\u0131m, ilerleyen d\u00f6nemde yerini niceliksel modellere b\u0131rakm\u0131\u015ft\u0131r. 20. y\u00fczy\u0131l\u0131n ikinci yar\u0131s\u0131nda standartla\u015fan\u00a0<strong>Lojistik Regresyon<\/strong>, kredi skorlamas\u0131nda (credit scoring) &#8220;alt\u0131n standart&#8221; haline gelmi\u015ftir. Lojistik regresyonun end\u00fcstri taraf\u0131ndan bu kadar benimsenmesinin temel nedeni, matematiksel basitli\u011fi ve y\u00fcksek yorumlanabilirli\u011fidir. Bir kredi tahsis uzman\u0131, modeldeki katsay\u0131lara bakarak, gelirin artmas\u0131n\u0131n temerr\u00fct riskini (Probability of Default &#8211; PD) ne kadar d\u00fc\u015f\u00fcrd\u00fc\u011f\u00fcn\u00fc kesin bir &#8220;olas\u0131l\u0131k oran\u0131&#8221; (odds ratio) ile ifade edebilmektedir.<\/p>\n<p>Dijital ekonominin y\u00fckseli\u015fiyle birlikte geleneksel modellerin s\u0131n\u0131rlar\u0131 zorlanmaya ba\u015flam\u0131\u015ft\u0131r. Lojistik regresyon, de\u011fi\u015fkenler aras\u0131nda do\u011frusal bir ili\u015fki varsayar (&#8220;bu m\u00fc\u015fteri krediyi \u00f6der mi, \u00f6demez mi?&#8221; gibi <strong>kategorik (evet\/hay\u0131r, 1\/0)<\/strong> sorulara yan\u0131t arar.) Oysa modern finansal davran\u0131\u015flar karma\u015f\u0131k ve \u00e7ok boyutludur. \u015e\u00f6yle ki; bir m\u00fc\u015fterinin temerr\u00fct riski sadece gelirine veya bor\u00e7 oran\u0131na ba\u011fl\u0131 de\u011fildir; harcama al\u0131\u015fkanl\u0131klar\u0131, sekt\u00f6rel d\u00f6ng\u00fcler, makroekonomik \u015foklar ve hatta dijital ayak izleri aras\u0131ndaki karma\u015f\u0131k etkile\u015fimlerden (interaction effects) vs. etkilenir. Geleneksel modeller, bu do\u011frusal olmayan \u00f6r\u00fcnt\u00fcleri (non-linear patterns) yakalamakta yetersiz kalmakta, bu da kredibilitesi olan m\u00fc\u015fteriyi reddetme (tip 1) ve riskli m\u00fc\u015fteriye kredi verme (tip 2) hatalar\u0131n\u0131n artmas\u0131na yol a\u00e7maktad\u0131r.<\/p>\n<p><strong>\u00a0\u0130statistiksel Temel: Lojistik Regresyonun Kal\u0131c\u0131l\u0131\u011f\u0131<\/strong><\/p>\n<p>Her ne kadar &#8220;eski&#8221; olarak nitelendirilse de, Lojistik Regresyon (LR) hala bir\u00e7ok bankan\u0131n, \u00f6zellikle sermaye yeterlili\u011fi (Basel Uzla\u015f\u0131lar\u0131) hesaplamalar\u0131nda kulland\u0131\u011f\u0131 temel modeldir. LR, ba\u011f\u0131ml\u0131 de\u011fi\u015fkenin (temerr\u00fct durumu: 0 veya 1) log-olas\u0131l\u0131\u011f\u0131n\u0131, ba\u011f\u0131ms\u0131z de\u011fi\u015fkenlerin do\u011frusal bir kombinasyonu olarak modeller.<\/p>\n<ul>\n<li><strong><em>G\u00fc\u00e7l\u00fc Y\u00f6nleri:<\/em><\/strong>\u00a0\u015eeffaft\u0131r, hesaplama maliyeti d\u00fc\u015f\u00fckt\u00fcr ve k\u00fc\u00e7\u00fck veri setlerinde bile kararl\u0131 (stable) sonu\u00e7lar verir. A\u015f\u0131r\u0131 \u00f6\u011frenme (overfitting) riski d\u00fc\u015f\u00fckt\u00fcr.<\/li>\n<li><strong><em>Zay\u0131f Y\u00f6nleri:<\/em><\/strong>\u00a0De\u011fi\u015fkenler aras\u0131ndaki karma\u015f\u0131k etkile\u015fimleri modelleyemez. \u00d6rne\u011fin, &#8220;Geliri y\u00fcksek ama ya\u015f\u0131 gen\u00e7&#8221; olan bir ki\u015finin riskini, bu iki de\u011fi\u015fkenin \u00e7arp\u0131m\u0131n\u0131 (interaction term) manuel olarak modele eklemeden yakalayamaz.<\/li>\n<\/ul>\n<p><strong>Makine \u00d6\u011frenimi: Karar A\u011fa\u00e7lar\u0131 ve Topluluk Y\u00f6ntemleri<\/strong><\/p>\n<p>Makine \u00f6\u011frenimi, verideki kal\u0131plar\u0131 otomatik olarak \u00f6\u011frenen algoritmalar sunar. Kredi riski y\u00f6netiminde en ba\u015far\u0131l\u0131 kategori, &#8220;Karar A\u011fa\u00e7lar\u0131&#8221; (Decision Trees) ve bunlar\u0131n birle\u015fiminden olu\u015fan &#8220;Topluluk \u00d6\u011frenme&#8221; (Ensemble Learning) y\u00f6ntemleridir.<\/p>\n<p><strong><em>\u00a0<\/em><\/strong><strong><em>Karar A\u011fa\u00e7lar\u0131 (Decision Trees)<\/em><\/strong><\/p>\n<p>Karar a\u011fa\u00e7lar\u0131, veri setini belirli kurallara g\u00f6re (\u00f6rne\u011fin &#8220;Gelir &gt; 50.000 TL mi?&#8221;) alt gruplara b\u00f6lerek bir a\u011fa\u00e7 yap\u0131s\u0131 olu\u015fturur.<\/p>\n<ul>\n<li><strong><em>Mekanizma:<\/em><\/strong><em>\u00a0<\/em>Algoritma, her ad\u0131mda &#8220;Bilgi Kazanc\u0131&#8221;n\u0131 (Information Gain) maksimize eden veya &#8220;Gini Safs\u0131zl\u0131\u011f\u0131&#8221;n\u0131 (Gini Impurity) minimize eden b\u00f6l\u00fcnmeyi se\u00e7er.<\/li>\n<li><strong><em>Risk:<\/em><\/strong>\u00a0Tek ba\u015f\u0131na kullan\u0131ld\u0131\u011f\u0131nda, e\u011fitim verisine a\u015f\u0131r\u0131 uyum sa\u011flama (overfitting) ve yeni verilerde ba\u015far\u0131s\u0131z olma e\u011filimindedir. Bu nedenle genellikle tek ba\u015f\u0131na kullan\u0131lmaz.<\/li>\n<\/ul>\n<p><strong><em>\u00a0<\/em><\/strong><strong><em>Random Forest (Rastgele Orman)<\/em><\/strong><\/p>\n<p>Random Forest, y\u00fczlerce karar a\u011fac\u0131n\u0131 bir araya getiren bir &#8220;Bagging&#8221; (Bootstrap Aggregating) tekni\u011fidir.<\/p>\n<ul>\n<li><strong><em>\u00c7al\u0131\u015fma Prensibi:<\/em><\/strong>\u00a0Her bir a\u011fa\u00e7, veri setinin rastgele se\u00e7ilmi\u015f bir alt k\u00fcmesiyle ve rastgele se\u00e7ilmi\u015f de\u011fi\u015fkenlerle e\u011fitilir. Nihai karar, t\u00fcm a\u011fa\u00e7lar\u0131n &#8220;oylamas\u0131&#8221; (s\u0131n\u0131fland\u0131rma) veya ortalamas\u0131 (regresyon) ile verilir.<\/li>\n<li><strong><em>Kredi Riskine Katk\u0131s\u0131:<\/em><\/strong>\u00a0Breiman (2001) taraf\u0131ndan geli\u015ftirilen bu y\u00f6ntem, a\u015f\u0131r\u0131 \u00f6\u011frenmeyi engeller ve y\u00fcksek boyutlu (\u00e7ok de\u011fi\u015fkenli) veri setlerinde m\u00fckemmel sonu\u00e7 verir. De\u011fi\u015fkenlerin \u00f6nem derecesini (feature importance) belirleyerek, hangi fakt\u00f6rlerin kredi riskini art\u0131rd\u0131\u011f\u0131n\u0131 g\u00f6sterir.\u00a0Ara\u015ft\u0131rmalar, Random Forest&#8217;\u0131n kredi risk tahmininde %93&#8217;e varan do\u011fruluk oranlar\u0131na ula\u015fabildi\u011fini raporlamaktad\u0131r.<\/li>\n<\/ul>\n<p><strong><em>\u00a0<\/em><\/strong><strong><em>Gradient Boosting Makineleri (GBM, XGBoost, LightGBM)<\/em><\/strong><\/p>\n<p>G\u00fcn\u00fcm\u00fczde kredi risk modellemede ve ileri d\u00fczey bankac\u0131l\u0131k uygulamalar\u0131nda en bask\u0131n model ailesi &#8220;Gradient Boosting&#8221;dir. Bu y\u00f6ntem, a\u011fa\u00e7lar\u0131 paralel de\u011fil, ard\u0131\u015f\u0131k (sequential) olarak in\u015fa eder. Her yeni a\u011fa\u00e7, bir \u00f6nceki a\u011fac\u0131n yapt\u0131\u011f\u0131 hatalar\u0131 (art\u0131klar\u0131 &#8211; residuals) d\u00fczeltmeye \u00e7al\u0131\u015f\u0131r.<\/p>\n<ul>\n<li><strong><em>XGBoost (Extreme Gradient Boosting):\u00a0<\/em><\/strong>H\u0131z, performans ve \u00f6l\u00e7eklenebilirlik a\u00e7\u0131s\u0131ndan end\u00fcstri standard\u0131 haline gelmi\u015ftir. Eksik verileri (missing values) otomatik olarak y\u00f6netebilir ve a\u015f\u0131r\u0131 \u00f6\u011frenmeyi engellemek i\u00e7in yerle\u015fik reg\u00fclarizasyon (L1\/L2) tekniklerine sahiptir.<\/li>\n<\/ul>\n<p>Yap\u0131lan kar\u015f\u0131la\u015ft\u0131rmal\u0131 \u00e7al\u0131\u015fmalarda, XGBoost&#8217;un %99,4 do\u011fruluk oran\u0131 ile Lojistik Regresyon, SVM ve Sinir A\u011flar\u0131n\u0131 geride b\u0131rakt\u0131\u011f\u0131 g\u00f6r\u00fclm\u00fc\u015ft\u00fcr.\u00a0Chang et al. (2022) \u00e7al\u0131\u015fmas\u0131nda XGBoost, %92,19 do\u011fruluk, 0,97 AUC ve %91,83 F1 skoru ile &#8220;en iyi genelleme yetene\u011fine sahip model&#8221; olarak tan\u0131mlanm\u0131\u015ft\u0131r.<\/p>\n<ul>\n<li><strong><em>LightGBM<\/em><\/strong>:\u00a0Microsoft taraf\u0131ndan geli\u015ftirilen bu algoritma, \u00f6zellikle \u00e7ok b\u00fcy\u00fck veri setlerinde XGBoost&#8217;tan daha h\u0131zl\u0131 e\u011fitim s\u00fcresi ve daha d\u00fc\u015f\u00fck bellek kullan\u0131m\u0131 sunar. Yaprak odakl\u0131 (leaf-wise) b\u00fcy\u00fcme stratejisi sayesinde derinlemesine \u00f6\u011frenir ve Asya pazarlar\u0131ndaki uygulamalarda y\u00fcksek F1 skorlar\u0131 elde etmi\u015ftir.<\/li>\n<li><strong><em>CatBoost:\u00a0<\/em><\/strong>Kategorik de\u011fi\u015fkenlerin (\u00f6rne\u011fin meslek grubu, \u015fehir) yo\u011fun oldu\u011fu veri setlerinde, \u00f6n i\u015fleme (one-hot encoding) gerektirmeden y\u00fcksek performans g\u00f6sterir.<\/li>\n<\/ul>\n<p><strong>Derin \u00d6\u011frenme (Deep Learning)<\/strong><\/p>\n<p>Yapay Sinir A\u011flar\u0131 (ANN) ve Derin Sinir A\u011flar\u0131 (DNN), insan beynindeki n\u00f6ronlar\u0131n \u00e7al\u0131\u015fma prensibini taklit eder.<\/p>\n<ul>\n<li><strong><em>Kullan\u0131m Alan\u0131:<\/em><\/strong><strong>\u00a0<\/strong>Yap\u0131land\u0131r\u0131lm\u0131\u015f (tablolama) kredi verilerinde Gradient Boosting modelleri genellikle daha iyi sonu\u00e7 verse de, Derin \u00d6\u011frenme \u00f6zellikle &#8220;yap\u0131land\u0131r\u0131lmam\u0131\u015f&#8221; verilerin (m\u00fc\u015fteri i\u015flem a\u00e7\u0131klamalar\u0131, ses kay\u0131tlar\u0131, belgeler) modele dahil edilmesinde kritiktir.<\/li>\n<li><strong><em>Hibrit Modeller:<\/em><\/strong>\u00a0Baz\u0131 ara\u015ft\u0131rmalar, XGBoost ile Sinir A\u011flar\u0131n\u0131n birle\u015ftirildi\u011fi (stacking) hibrit yap\u0131lar\u0131n, tekil modellerden daha kararl\u0131 sonu\u00e7lar verdi\u011fini g\u00f6stermektedir.<\/li>\n<\/ul>\n<p><strong>\u00a0<\/strong><strong>Destek Vekt\u00f6r Makineleri (SVM)<\/strong><\/p>\n<p>SVM, veriyi s\u0131n\u0131flara ay\u0131rmak i\u00e7in en uygun hiper d\u00fczlemi bulmaya \u00e7al\u0131\u015f\u0131r. K\u00fc\u00e7\u00fck veri setlerinde y\u00fcksek do\u011fruluk sa\u011flasa da, b\u00fcy\u00fck bankac\u0131l\u0131k verilerinde hesaplama maliyeti \u00e7ok y\u00fcksektir ve \u00f6l\u00e7eklenmesi zordur. Bu nedenle modern kredi skorlamas\u0131nda daha az tercih edilmektedir.<\/p>\n<table>\n<tbody>\n<tr>\n<td>&nbsp;<\/p>\n<p><strong>\u00d6zellik<\/strong><\/td>\n<td><strong>Lojistik Regresyon<\/strong><\/td>\n<td><strong>Random Forest<\/strong><\/td>\n<td><strong>XGBoost \/ LightGBM<\/strong><\/td>\n<td><strong>Derin \u00d6\u011frenme (DL)<\/strong><\/td>\n<\/tr>\n<tr>\n<td><strong>Tahmin G\u00fcc\u00fc<\/strong><\/td>\n<td>D\u00fc\u015f\u00fck\/Orta<\/td>\n<td>Y\u00fcksek<\/td>\n<td>\u00c7ok Y\u00fcksek<\/td>\n<td>Y\u00fcksek (Veriye Ba\u011fl\u0131)<\/td>\n<\/tr>\n<tr>\n<td><strong>Yorumlanabilirlik<\/strong><\/td>\n<td>\u00c7ok Y\u00fcksek (\u015eeffaf)<\/td>\n<td>Orta (Feature Imp.)<\/td>\n<td>D\u00fc\u015f\u00fck (XAI Gerekli)<\/td>\n<td>\u00c7ok D\u00fc\u015f\u00fck (Kara Kutu)<\/td>\n<\/tr>\n<tr>\n<td><strong>Veri Boyutu<\/strong><\/td>\n<td>K\u00fc\u00e7\u00fck\/Orta<\/td>\n<td>B\u00fcy\u00fck<\/td>\n<td>\u00c7ok B\u00fcy\u00fck<\/td>\n<td>Devasa<\/td>\n<\/tr>\n<tr>\n<td><strong>E\u011fitim H\u0131z\u0131<\/strong><\/td>\n<td>\u00c7ok H\u0131zl\u0131<\/td>\n<td>Orta<\/td>\n<td>H\u0131zl\u0131<\/td>\n<td>Yava\u015f<\/td>\n<\/tr>\n<tr>\n<td><strong>A\u015f\u0131r\u0131 \u00d6\u011frenme Riski<\/strong><\/td>\n<td>D\u00fc\u015f\u00fck<\/td>\n<td>D\u00fc\u015f\u00fck<\/td>\n<td>Orta (Ayar Gerektirir)<\/td>\n<td>Y\u00fcksek<\/td>\n<\/tr>\n<tr>\n<td><strong>End\u00fcstriyel Kullan\u0131m<\/strong><\/td>\n<td>Y\u00fcksek (D\u00fczenleyici)<\/td>\n<td>Y\u00fcksek<\/td>\n<td>\u00c7ok Y\u00fcksek (Karar)<\/td>\n<td>D\u00fc\u015f\u00fck\/Orta (Fraud\/G\u00f6r\u00fcnt\u00fc)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><em>\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0Kredi Risk Y\u00f6netiminde Kullan\u0131lan Temel Algoritmalar\u0131n Kar\u015f\u0131la\u015ft\u0131rmal\u0131 Analizi<\/em><\/p>\n<p><strong>Kara Kutu Sorunu ve A\u00e7\u0131klanabilir Yapay Zeka (XAI)<\/strong><\/p>\n<p>Makine \u00f6\u011frenimi modellerinin artan karma\u015f\u0131kl\u0131\u011f\u0131, beraberinde ciddi bir sorunu getirmi\u015ftir:\u00a0<strong>\u015eeffafl\u0131k Eksikli\u011fi<\/strong>. S\u00f6z gelimi; bir bankan\u0131n m\u00fc\u015fterisine &#8220;krediniz reddedildi \u00e7\u00fcnk\u00fc modelimiz 0.45 skor \u00fcretti&#8221; demesi ne m\u00fc\u015fteri memnuniyeti a\u00e7\u0131s\u0131ndan ne de yasal d\u00fczenlemeler a\u00e7\u0131s\u0131ndan kabul edilebilirdir. BDDK, Avrupa Merkez Bankas\u0131 (ECB) ve di\u011fer d\u00fczenleyiciler, kararlar\u0131n\u00a0<em>neden<\/em>\u00a0al\u0131nd\u0131\u011f\u0131n\u0131n a\u00e7\u0131klanmas\u0131n\u0131 zorunlu k\u0131lar. Bu ihtiya\u00e7,\u00a0<strong>A\u00e7\u0131klanabilir Yapay Zeka (Explainable AI &#8211; XAI)<\/strong>\u00a0disiplinini finans\u0131n merkezine ta\u015f\u0131m\u0131\u015ft\u0131r.<\/p>\n<p>Geleneksel g\u00f6r\u00fc\u015f, modelin performans\u0131 (do\u011frulu\u011fu) artt\u0131k\u00e7a a\u00e7\u0131klanabilirli\u011finin azald\u0131\u011f\u0131 y\u00f6n\u00fcndedir. Lojistik regresyon tamamen \u015feffafken, binlerce a\u011fa\u00e7tan olu\u015fan bir XGBoost modeli veya milyonlarca parametreli bir Sinir A\u011f\u0131, insan alg\u0131s\u0131n\u0131n \u00f6tesinde bir &#8220;Kara Kutu&#8221;dur. XAI teknikleri, bu kutuyu a\u00e7arak modelin i\u00e7 i\u015fleyi\u015fini veya en az\u0131ndan kararlar\u0131n\u0131n gerek\u00e7elerini g\u00f6r\u00fcn\u00fcr k\u0131lar.<\/p>\n<p><strong>Veri Devrimi: Alternatif Veri ve \u00d6zellik M\u00fchendisli\u011fi<\/strong><\/p>\n<p>Geleneksel kredi skorlamas\u0131, b\u00fcy\u00fck \u00f6l\u00e7\u00fcde ge\u00e7mi\u015f finansal i\u015flem kay\u0131tlar\u0131na (kredi kart\u0131 \u00f6demeleri, kredi taksitleri) ve kredi b\u00fcrosu (KKB, Findecks, Experian) verilerine dayan\u0131r. Bu yakla\u015f\u0131m, finansal ge\u00e7mi\u015fi olmayan (thin-file) veya bankac\u0131l\u0131k sistemi d\u0131\u015f\u0131nda kalan (unbanked) milyarlarca insan\u0131 sistemin d\u0131\u015f\u0131nda b\u0131rakmak gibi problemleri i\u00e7erir. \u0130\u015fte yapay zeka,\u00a0alternatif veri\u00a0kaynaklar\u0131n\u0131 i\u015fleyerek bu &#8220;g\u00f6r\u00fcnmez&#8221; riski \u00f6l\u00e7\u00fclebilir hale getirir.<\/p>\n<p><strong><em>Dijital Ayak \u0130zleri ve Telekom Verileri<\/em><\/strong><\/p>\n<p>Ak\u0131ll\u0131 telefon kullan\u0131m verileri, bir ki\u015finin finansal disiplini ve ya\u015fam tarz\u0131 hakk\u0131nda g\u00fc\u00e7l\u00fc sinyaller verir.<\/p>\n<ul>\n<li><strong><em>Vaka (Filipinler):<\/em><\/strong>Tonik Bank, FinScore&#8217;un yapay zeka destekli &#8220;Telco Data Credit Scoring&#8221; mod\u00fcl\u00fcn\u00fc kullanarak, ak\u0131ll\u0131 telefon penetrasyonunun y\u00fcksek ancak bankac\u0131l\u0131\u011f\u0131n d\u00fc\u015f\u00fck oldu\u011fu Filipinler pazar\u0131nda kredi eri\u015fimini art\u0131rm\u0131\u015ft\u0131r.<\/li>\n<\/ul>\n<p><strong><em>E-Ticaret ve Davran\u0131\u015fsal Analitik<\/em><\/strong><\/p>\n<p>KOB\u0130 kredilerinde, i\u015fletmenin e-ticaret platformundaki performans\u0131 (sat\u0131\u015f hacmi, m\u00fc\u015fteri yorumlar\u0131, iade oranlar\u0131, stok devir h\u0131z\u0131) geleneksel bilan\u00e7olardan \u00e7ok daha g\u00fcncel bir risk g\u00f6stergesidir. Bir \u00e7al\u0131\u015fma, bir e-ticaret firmas\u0131n\u0131n dijital ayak izlerini kullanan modellerin, temerr\u00fct oranlar\u0131n\u0131 yakla\u015f\u0131k %30 oran\u0131nda azaltt\u0131\u011f\u0131n\u0131 ve i\u015fletme kar marj\u0131n\u0131 %10 art\u0131rd\u0131\u011f\u0131n\u0131 ortaya koymu\u015ftur.<\/p>\n<p><strong><em>\u00a0Sosyal ve Psikometrik Veriler<\/em><\/strong><\/p>\n<p>Daha tart\u0131\u015fmal\u0131 bir alan olmakla birlikte, sosyal medya verileri (ba\u011flant\u0131lar\u0131n kalitesi, e\u011fitim durumu &#8211; LinkedIn) ve psikometrik testler (ki\u015filik envanterleri) de risk modellemesinde kullan\u0131lmaktad\u0131r.<\/p>\n<p>Bu t\u00fcr verilerin kullan\u0131m\u0131, etik kayg\u0131lar\u0131 ve veri gizlili\u011fi (privacy) tart\u0131\u015fmalar\u0131n\u0131 beraberinde getirir. \u00d6zellikle T\u00fcrkiye&#8217;de KVKK ve Avrupa&#8217;da GDPR, bu verilerin izinsiz kullan\u0131m\u0131n\u0131 kesinlikle yasaklar. Ancak kullan\u0131c\u0131n\u0131n a\u00e7\u0131k r\u0131zas\u0131 ile (Open Banking \/ A\u00e7\u0131k Bankac\u0131l\u0131k kapsam\u0131nda) bu veriler zenginle\u015ftirici fakt\u00f6r olarak modele dahil edilebilir.<\/p>\n<p><strong>D\u00fczenleyici \u00c7er\u00e7eve: \u0130novasyon ve Kontrol Aras\u0131ndaki Denge<\/strong><\/p>\n<p>Yapay zekan\u0131n finansal kararlarda kullan\u0131m\u0131, d\u00fcnya genelinde d\u00fczenleyicilerin merce\u011fi alt\u0131ndad\u0131r. Hatal\u0131, \u00f6nyarg\u0131l\u0131 veya a\u00e7\u0131klanamayan bir model, sistemik risklere yol a\u00e7abilir.<\/p>\n<p><strong><em>Basel IV ve Sermaye Gereksinimleri<\/em><\/strong><\/p>\n<p>Bankac\u0131l\u0131k d\u00fczenlemelerinin k\u00fcresel standard\u0131 olan Basel Uzla\u015f\u0131s\u0131 (Basel III\/IV), bankalar\u0131n kredi riski modellerini (IRB &#8211; Internal Ratings-Based) sermaye hesaplamas\u0131nda kullanmas\u0131na izin verir ancak kat\u0131 kurallar koyar.<\/p>\n<p><strong><em>AB Yapay Zeka Yasas\u0131 (EU AI Act)<\/em><\/strong><\/p>\n<p>Avrupa Birli\u011fi&#8217;nin kabul etti\u011fi ve k\u00fcresel bir standart olma yolunda ilerleyen Yapay Zeka Yasas\u0131, kredi skorlama sistemlerini\u00a0<strong>&#8220;Y\u00fcksek Riskli&#8221;<\/strong>\u00a0(High Risk) kategorisine alm\u0131\u015ft\u0131r.<\/p>\n<p><strong>T\u00fcrkiye&#8217;deki D\u00fczenleyici \u00c7er\u00e7eve<\/strong><\/p>\n<p>T\u00fcrkiye&#8217;de Bankac\u0131l\u0131k D\u00fczenleme ve Denetleme Kurumu (BDDK) ve T\u00fcrkiye Cumhuriyet Merkez Bankas\u0131 (TCMB), teknolojiyi yak\u0131ndan takip eden proaktif bir tutum sergilemektedir.<\/p>\n<ul>\n<li><strong><em>Bilgi Sistemleri Y\u00f6netmeli\u011fi:<\/em><\/strong>\u00a0Bankalar\u0131n kulland\u0131\u011f\u0131 risk modellerinin validasyonu, testi ve y\u00f6neti\u015fimi konusunda s\u0131k\u0131 kurallar mevcuttur. Yapay zeka modelleri de bu y\u00f6netmeliklere tabidir.<\/li>\n<li><strong><em>Fintech Denetimleri:<\/em><\/strong>\u00a0TCMB, \u00f6deme ve elektronik para kurulu\u015flar\u0131n\u0131 s\u0131k\u0131 bir denetim alt\u0131nda tutmaktad\u0131r. \u00d6rne\u011fin, 2025 y\u0131l\u0131nda Sipay ve Vepara gibi kurulu\u015flar\u0131n faaliyet izinlerinin ge\u00e7ici olarak durdurulmas\u0131, uyum (compliance) s\u00fcre\u00e7lerinin ve teknolojik altyap\u0131 g\u00fcvenli\u011finin ne kadar kritik oldu\u011funu g\u00f6stermi\u015ftir.<\/li>\n<li><strong><em>Etik \u0130lkeler:<\/em><\/strong>\u00a0T\u00fcrkiye&#8217;de hen\u00fcz AB AI Act benzeri tekil bir yasa olmasa da, akademik ve sekt\u00f6rel \u00e7al\u0131\u015fmalar &#8220;Yapay Zeka Eti\u011fi&#8221; (\u015feffafl\u0131k, hesap verebilirlik, adalet) ilkelerinin finans sekt\u00f6r\u00fcne entegrasyonuna odaklanmaktad\u0131r. KVKK, ki\u015fisel verilerin (\u00f6zellikle alternatif verilerin) i\u015flenmesinde ana d\u00fczenleyici metindir.<\/li>\n<\/ul>\n<p><strong>\u00a0<\/strong><strong>Sonu\u00e7<\/strong><\/p>\n<p>Kredi risk y\u00f6netiminde yapay zeka ve makine \u00f6\u011frenimi uygulamalar\u0131, finans sekt\u00f6r\u00fcnde geri d\u00f6nd\u00fcr\u00fclemez bir paradigma de\u011fi\u015fimini temsil etmektedir. Ara\u015ft\u0131rmam\u0131z, bu teknolojilerin sadece &#8220;daha h\u0131zl\u0131 i\u015flem yapmak&#8221; i\u00e7in de\u011fil, &#8220;riski daha iyi anlamak&#8221; ve &#8220;finansal eri\u015fimi demokratikle\u015ftirmek&#8221; i\u00e7in kritik bir ara\u00e7 oldu\u011funu net bir bi\u00e7imde ortaya koymaktad\u0131r.<\/p>\n<p>Geleneksel lojistik regresyon modellerinden XGBoost ve Derin \u00d6\u011frenme gibi geli\u015fmi\u015f algoritmalara ge\u00e7i\u015f, tahmin do\u011frulu\u011funda devrimsel etki yaratm\u0131\u015ft\u0131r. Ancak bu devrim, &#8220;Kara Kutu&#8221; sorununu beraberinde getirmi\u015f ve XAI teknolojilerini ka\u00e7\u0131n\u0131lmaz bir zorunluluk k\u0131lm\u0131\u015ft\u0131r. \u00dcretken yapay zeka ve ajanik sistemler ise bu d\u00f6n\u00fc\u015f\u00fcm\u00fc bir ad\u0131m \u00f6teye ta\u015f\u0131yarak, bankac\u0131l\u0131\u011f\u0131 reaktif bir yap\u0131dan proaktif ve otonom bir yap\u0131ya d\u00f6n\u00fc\u015ft\u00fcrmektedir.<\/p>\n<p>\u00dclkemiz \u00f6zelinde bakt\u0131\u011f\u0131m\u0131zda ise, T\u00fcrkiye\u2019nin g\u00fc\u00e7l\u00fc bankac\u0131l\u0131k altyap\u0131s\u0131 ve dinamik fintech ekosistemiyle bu d\u00f6n\u00fc\u015f\u00fcme haz\u0131r oldu\u011fu rahatl\u0131kla s\u00f6ylenebilir. Ancak ba\u015far\u0131n\u0131n anahtar\u0131; veriyi etik, \u015feffaf, g\u00fcvenli ve d\u00fczenlemelere tam uyumlu bir \u015fekilde i\u015fleyebilen, teknolojiyi insan uzmanl\u0131\u011f\u0131yla harmanlayan b\u00fct\u00fcnc\u00fcl bir risk y\u00f6netimi k\u00fclt\u00fcr\u00fcn\u00fc in\u015fa etmekten ge\u00e7mektedir.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Kredi risk y\u00f6netiminde yapay zeka ve makine \u00f6\u011frenimi\u00a0uygulamalar\u0131, bor\u00e7lananlar\u0131n risk durumlar\u0131n\u0131 \u00f6l\u00e7mek i\u00e7in kullan\u0131lan yeni jenerasyon finansal ara\u00e7lard\u0131r.\u00a0Kredi riski, bor\u00e7lunun borcunu \u00f6deyememe olas\u0131l\u0131\u011f\u0131 olarak tan\u0131mlan\u0131r ve riskin y\u00f6netimi, finansal sistemin temel dinami\u011fini olu\u015fturan kaynak fazlas\u0131 olan birimlerden kaynak a\u00e7\u0131\u011f\u0131 olan birimlere fon ak\u0131\u015flar\u0131n\u0131n etkinli\u011fi a\u00e7\u0131s\u0131ndan hem bireysel (t\u00fcketici) hem de kurumsal (ticari) kredilerde ileri d\u00fczey<\/p>\n","protected":false},"author":16,"featured_media":5629,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[40,6,520],"tags":[544,537,170],"class_list":{"0":"post-5628","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-finansal-otomasyon","8":"category-finansal-veri-okuryazarligi","9":"category-yapay-zeka","10":"tag-kredi-skorlamasi","11":"tag-makine-ogrenmesi","12":"tag-yapay-zeka"},"better_featured_image":{"id":5629,"alt_text":"Kredi ve Risk Y\u00f6netiminde Makine \u00d6\u011frenimi Uygulamalar\u0131","caption":"","description":"","media_type":"image","media_details":{"width":2560,"height":1708,"file":"2025\/12\/Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-scaled.jpg","filesize":386874,"sizes":{"medium":{"file":"Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-300x200.jpg","width":300,"height":200,"mime-type":"image\/jpeg","filesize":20147,"source_url":"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2025\/12\/Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-300x200.jpg"},"large":{"file":"Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-1024x683.jpg","width":1024,"height":683,"mime-type":"image\/jpeg","filesize":101777,"source_url":"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2025\/12\/Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-1024x683.jpg"},"thumbnail":{"file":"Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-150x150.jpg","width":150,"height":150,"mime-type":"image\/jpeg","filesize":11968,"source_url":"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2025\/12\/Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-150x150.jpg"},"medium_large":{"file":"Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-768x512.jpg","width":768,"height":512,"mime-type":"image\/jpeg","filesize":68335,"source_url":"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2025\/12\/Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-768x512.jpg"},"1536x1536":{"file":"Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-1536x1025.jpg","width":1536,"height":1025,"mime-type":"image\/jpeg","filesize":180254,"source_url":"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2025\/12\/Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-1536x1025.jpg"},"2048x2048":{"file":"Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-2048x1366.jpg","width":2048,"height":1366,"mime-type":"image\/jpeg","filesize":275245,"source_url":"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2025\/12\/Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-2048x1366.jpg"},"bunyad-small":{"file":"Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-150x100.jpg","width":150,"height":100,"mime-type":"image\/jpeg","filesize":9181,"source_url":"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2025\/12\/Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-150x100.jpg"},"bunyad-medium":{"file":"Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-450x300.jpg","width":450,"height":300,"mime-type":"image\/jpeg","filesize":33931,"source_url":"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2025\/12\/Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-450x300.jpg"},"bunyad-full":{"file":"Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-1200x801.jpg","width":1200,"height":801,"mime-type":"image\/jpeg","filesize":127184,"source_url":"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2025\/12\/Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-1200x801.jpg"},"bunyad-viewport":{"file":"Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-2048x1366.jpg","width":2048,"height":1366,"mime-type":"image\/jpeg","filesize":275245,"source_url":"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2025\/12\/Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-2048x1366.jpg"},"bunyad-768":{"file":"Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-768x512.jpg","width":768,"height":512,"mime-type":"image\/jpeg","filesize":68335,"source_url":"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2025\/12\/Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-768x512.jpg"}},"image_meta":{"aperture":"0","credit":"","camera":"","caption":"","created_timestamp":"0","copyright":"","focal_length":"0","iso":"0","shutter_speed":"0","title":"","orientation":"0","keywords":[]},"original_image":"Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari.jpg"},"post":5628,"source_url":"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2025\/12\/Kredi-ve-Risk-Yonetiminde-Yapay-Zeka-Uygulamalari-scaled.jpg"},"amp_enabled":true,"_links":{"self":[{"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/posts\/5628","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/users\/16"}],"replies":[{"embeddable":true,"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/comments?post=5628"}],"version-history":[{"count":2,"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/posts\/5628\/revisions"}],"predecessor-version":[{"id":5631,"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/posts\/5628\/revisions\/5631"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/media\/5629"}],"wp:attachment":[{"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/media?parent=5628"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/categories?post=5628"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/tags?post=5628"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}