ATM nakit ikmal optimizasyonunda asimetrik destek vektör regresyon tahmin modeli yaklaşımı
Asymmetric support vector regression forecast model approach in ATM cash replenishment optimization
- Tez No: 439456
- Danışmanlar: YRD. DOÇ. DR. CAFER ERHAN BOZDAĞ
- Tez Türü: Doktora
- Konular: Endüstri ve Endüstri Mühendisliği, Industrial and Industrial Engineering
- Anahtar Kelimeler: Belirtilmemiş.
- Yıl: 2016
- Dil: Türkçe
- Üniversite: İstanbul Teknik Üniversitesi
- Enstitü: Fen Bilimleri Enstitüsü
- Ana Bilim Dalı: Endüstri Mühendisliği Ana Bilim Dalı
- Bilim Dalı: Belirtilmemiş.
- Sayfa Sayısı: 162
Özet
ATM nakit tahmini ve nakit ikmali problemi bir başka deyişle ATM nakit yönetimi bankalar için büyük bir maliyet unsurudur. Banka müşterilerinin memnuniyetinin sağlanması ve bir yandan da bankanın nakit israfının önüne geçilmesi için hem yeterli performansı gösterecek bir tahmin modelinin geliştirilmesi hem de optimum nakit ikmal planlamasının yapılabilmesi gerekmektedir. Bugüne kadar yapılmış çalışmalarda bu iki durum ayrı ayrı ele alınmış olup; nakit tahmini ile nakit ikmalini entegre bir şekilde gerçekleştiren kapsamlı bir çalışma bulunmamaktadır. Bu çalışmada, yeni asimetrik tahmin modelleri ve bu modeller ile entegre olarak çalışan ve nakit ikmal maliyetlerini minimize eden bir optimizasyon modeli önerilmiştir. Uzman görüşüne dayanarak kullanılan değişkenler ve yeni önerilen veri düzenleme yöntemleri ile ATM nakit yönetimi için önerilen optimizasyon modeli hem önerilen hem de mevcut tahmin yöntemleri ile birlikte çalıştırılmış ve bu modelden alınan nakit ikmal kararlarına göre tüm tahmin modellerinin sonuçları karşılaştırılmıştır. Bu tezde önerilen ve literatürde mevcut bulunan tahmin modelleri, optimizasyon modelleri, algoritmaları içeren tüm kodlamalar R istatistiksel programlama dili ve R 3.2.2 yazılım versiyonu kullanılarak geliştirilmiştir. Önerilen asimetrik hata varyansını minimize eden destek vektör regresyon modeli ile diğer tahmin modellerine kıyasla daha düşük nakit yönetim maliyetleri elde edildiği deneysel çalışmalar ile gösterilmiştir. Aynı zamanda her gün nakit ikmalinin yapılmadığı durumlarda günlük nakit ikmal optimizasyon sonuçlarına göre kümülatif nakit ikmal optimizasyonu sonuçlarının daha az nakit ile ve daha düşük maliyetle gerçekleştirilebildiği gösterilmiştir. Hem yeni bir tahmin modeli önerisi hem de yeni bir nakit ikmal optimizasyon modeli yaklaşımı ve bu modellerin entegre bir şekilde kullanımını sağlayan bir algoritma önerilmesi nedeniyle literatüre özgün bir çalışma sunulmuştur. Aynı zamanda ATM nakit ikmal maliyetlerine Gezgin Satıcı Problemi algoritması kullanılarak elde edilen ulaşım maliyetleri de eklenerek, önerilen modelin maliyet avantajını koruduğu gözlemlenmiştir. Son olarak önerilen ve mevcut tahmin modelleri için çok kriterli karar verme yöntemi olan TOPSIS yöntemi ile maliyet, hata büyüklükleri, müşteri talebini karşılama oranı, ATM'lere nakit yükleme miktarları göz önünde bulundurularak genel bir sıralama yapılmıştır. TOPSIS yöntemi ile elde edilen sıralamada önerilen asimetrik destek vektör regresyon tahmin modeli birinci sırada, önerilen kümülatif optimum yükleme frekanslı destek vektör regresyon modeli ikinci sırada ve V-TUBE modeli üçüncü sırada yer almıştır. Tezin birinci bölümünde tezin amacının açıklanmasına yönelik genel bir giriş yapılmış ve ATM nakit yönetimi süreci hakkında bilgi verilmiştir. Tezin ikinci bölümünde literatürde mevcut olan ATM nakit tahmin ve optimizasyon modelleri hakkında bilgiler verilmiş ve incelenen tahmin modelleri açıklanmıştır. Üçüncü bölümde ATM nakit yönetimi için kullanılan verilerin analizi ve veri iyileştirme yöntemleri ile önerilen tahmin modelleri, optimizasyon modeli ve algoritması ile ATM'ler için rotalama ve modellerin sıralanması ve değerlendirilmesi amacı için kullanılan çok kriterli karar verme yöntemi için açıklamalar yer almaktadır. Tezin dördüncü bölümünde önerilen ve literatürdeki mevcut tahmin modellerinin karşılaştırılabilmesi için yöntemlerin deneysel uygulama sonuçları yer almaktadır. Tezin son bölümü olan beşinci bölümde ise tüm uygulama sonuçları değerlendirilmektedir.
Özet (Çeviri)
The costs in banking and finance sector constitute importance especially in different distribution channels such as automatic teller machines (ATM's), internet banking, mobile banking, branches, and call centers. Among these, ATM cash demand forecasting, cash supply, cash optimization and ATM routing problem, in other words, ATM cash management has a special importance since it is a major cost factor in many ways. In order to avoid large costs, Banks seek new solution methods. While there is a need for meeting customer demand, avoiding wasting cash require for optimum cash supply planning and developing a robust cash forecast model. Here also understood that, as well as choosing the forecast model giving the smallest forecast error ratio for the solution of this problem, finding the optimization model giving the minimum replenishment costs has importance. The problem of when and how much money should be replenished in each ATM and which routing should be used ought to be solved. Also, ATM cash management performance criteria should be evaluated, such as cash out ratio, amount of money loaded to ATM, frequency of cash loading, overall cash management costs, cash demand forecast error ratios. There are two aspects in ATM cash management costs. One of the two is the service provider or Banks and the other one is the customer. These aspects should be handled establishing a balance between the two, according to expert opinion. In studies so far, ATM cash forecasting and cash supply are considered separately and thus, there is no study integrating the replenishment optimization and forecast part of ATM cash management. There have been assumptions such as decreasing the cash demand forecasting errors provides decreasing the cash management costs or grouping and aggregating the ATM's for forecasting and then converting group forecast results to individual forecasts helps to obtain reduced cash management results. Some of the studies only consider replenishment costs without forecasts and some of them only focus on forecasting methods of ATM cash demand. In addition, some of them handles inventory routing problems of ATM's with proviously known customer demand. In addition, in previous studies, special days, which affect the amount of money withdrawn, were not analyzed with details and ony indiator variables are used in forecast models. Special days are handled as holidays or national days in previous studies. Whereas, especially in banking sector, other special variables such as employee and retired salary payment days and periodic payment days might affect the amount of money withdrawn. In this study, three new asymmetrical forecast models and an optimization model running integrated with these forecast models, in other words, an optimization model that reduces the two stage of forecast and optimization process to a single stage is proposed. At the same time, a comparison of costs with the other forecast models in the literature is performed. The main question or perspective for developing forecast models was that if controlling the variability of forecast errors besides the magnitude of the errors helped to reduce the ATM cash management costs and differed from other standard error minimizing forecast methods. In addition, another motivation for this study was that if the parameters of the forecasting models could be arranged according to a determined goal and not only for minimizing demand forecast errors. Another crucial point in this study was to find the answer of the question of how to find the amount of money replaced and when to replace money in each ATM without passing through many stages and assumptions such as known demand or known cash loading periodicity or known ATM routing. The contents of this thesis is explained chapter by chapter. In the introduction part of this thesis, the need for the proper solution of the ATM cash management was emphasized and the shortcomings of the existing methods are explained. Also, the reason for selecting the proposed methods and material was announced. In addition, an overall introduction for the next sections was given. Also, ATM cash management process was explained and the real world problems were revealed. In the second part of the thesis, existing methods for ATM cash forecasting, optimization methods, and their shortcomings were explained. Also, the difference between existing ATM cash management and forecasting methods and the proposed forecast and optimization method was revealed. In addition, existing forecast methods in the literature were explained in order to compare them with proposed forecasting methods and the formulation for these methods were given. The third part of the thesis consists the details of the proposed methods. First, ATM data were analyzed and proposed data correction methods were introduced. The selected variables according to financial expert opinion were explained. For data correction, new data correction algorithms for salary payment, feast and other public holidays were introduced. In this section, instead of using indicator or dummy variables, which are insuffient for the case of diversity of special days, the place of these special days were changed so that the input for special days of the output variables were arranged in a way to overcome day shifts of lagged variables in special days. In addition, R codes for data correction algorithms were explained. In addition, three new asymmetric forecast models and R programs, which are used for coding proposed new forecast models, were introduced. Optimization model and ATM cash replenishment algorithms are also introduced in the fifth part. Additionally, Traveling Salesman Problem for routing of cash and TOPSIS method for evaluating forecast models were explained. The fourth part of the thesis gives the experimental optimization results of the proposed and existing forecast models for both validation and test data. For optimization of forecast model parameters, both Nelder-Mead and Broyden-Fletcher-Goldfarb-Shanno Algoritm were used in validation data. The paramters of the method giving the minimum cost in the validation data were used in test data. All forecast models, optimization model and algorithms were coded in version 3.2.2 of R Statistical Program. According to the results of when to replenish and how much money to upload for each ATM, Traveling Salesman Problem is solved in order to complete the cash uplading route in a shortest path. In addition, a multi criteria decision making method, TOPSIS is used for overal evaluation of forecast methods. Proposed asymmetrical variance minimizing support vector regression forecast model was found to give the smallest cost of cash management according to other forecasting models. Also, secondly proposed asymetrical support vector regression model, which is called cumulative optimum replenishment frequency asymmetric variance minimizing support vector regression model was found to give the third smallest cost of cash replenisment. It is denoted that the proposed model had smaller symetric mean absolute percentage error than v-tube asymmetrical support vector regression, of which the cost of cash replenisment is the same. The thirdly proposed asymmetrical support vector regression is different from cumulative optimum replenishment frequency asymmetric variance minimizing support vector regression model. This forecast model assumes a constant replenishment period to be optimized differently from cumulative optimum replenishment frequency asymmetric variance minimizing support vector regression model. In cumulative optimum replenishment frequency asymmetric variance minimizing support vector regression model, optimum replenishment frequency is determined in the validation process. One interesting result is that, inspite of optimizing cumulative cash withdrawal results for a constant period, the thirdly proposed asymmetrical forecast model had the smallest forecast errors compared to other asymmetrical forecast models. The last part of this study summarizes the experimental results of the fourth part and gives an overal evaluation. The contribution of this thesis to the literature is that, this study integrates the forecasting and optimization process in one algorithm in order to decide how much money to replace and when to replace money in one stage and for each ATM individually. Furthermore, the routing problem is solved according to the optimized replenishment frequencies of each ATM. In combination with the innovation in ATM cash management process, three new forecasting models which, have different advantages have been proposed. Until this time, there have been studies for ATM cash forecasting, replenisment and routing, but these studies were not intended as comprehensive and detailed study as this study.
Benzer Tezler
- Comparison of integer linear programming and dynamic programming approaches for atm cash replenishment optimization problem
Atm nakit yenileme optimizasyonu problemi için tamsayılı doğrusal programlama ve dinamik programlama yaklaşımlarının karşılaştırılması
FAZİLET ÖZER
Yüksek Lisans
İngilizce
2019
BankacılıkOrta Doğu Teknik ÜniversitesiBilgisayar Mühendisliği Ana Bilim Dalı
PROF. DR. İSMAİL HAKKI TOROSLU
PROF. DR. PINAR KARAGÖZ
- Cash replenishment strategies and vehicle routing problem for automated teller machines
Otomatik vezne makineleri için araç rotalama problemi ve nakit ikmal stratejileri
DENİZ ORHAN
Yüksek Lisans
İngilizce
2023
Endüstri ve Endüstri MühendisliğiGalatasaray ÜniversitesiEndüstri Mühendisliği Ana Bilim Dalı
PROF. DR. MÜJDE GENEVOIS
- Otomatik para çekme makinelerinde (ATM) nakit para yönetimi için optimizasyon yaklaşımlarının geliştirilmesi
Development of optimization approaches for cash management in automatic teller machi̇nes (ATM)
ALİ TUNÇ
Doktora
Türkçe
2023
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolSelçuk ÜniversitesiBilgisayar Mühendisliği Ana Bilim Dalı
PROF. DR. ŞAKİR TAŞDEMİR
- A comprehensive decision-making approach to the atm cashflow management
Bankamatiklerdeki nakit akışına yönelik bütünsel kararverme yaklaşımı
MİCHELE CEDOLİN
Doktora
İngilizce
2022
Endüstri ve Endüstri MühendisliğiGalatasaray ÜniversitesiEndüstri Mühendisliği Ana Bilim Dalı
PROF. DR. MÜJDE GENEVOIS
- Nakit taşıma, sayım, saklama hizmetlerinde dış kaynak kullanımı
Outsourching for cash in transit sector
MEHMET TECER
Yüksek Lisans
Türkçe
2017
BankacılıkBeykent Üniversitesiİşletme Yönetimi Ana Bilim Dalı
YRD. DOÇ. DR. İSMAİL ERKAN ÇELİK