Bankamatiklerdeki nakit akışına yönelik bütünsel kararverme yaklaşımı

ATM cash flow optimization problem has two conflicting objectives. The first objective is to minimize costs. Money stocked in ATMs does not profit from overnight interest. However, insufficient replenishment results with customer dissatisfaction. The second objective being to maximize the service level, a balance should therefore be found between the costs and the service levels. In this thesis, the literature on ATM demand forecasting was reviewed, a common data set was found and the results from the literature are observed. First, statistical estimation techniques; ARIMA and SARIMA, then machine learning techniques; DNN, RNN and LSTM were used and outperforming results were obtained. To our knowledge, RNN and LSTM models were used for the first time in this field. In the second step, the estimated demand was used as the input to the replenishment problem. In the application, two static inventory control mechanisms have been used; the economic order quantity, in which ATMs are filled with fixed quantities on each visit, and lot for lot replenishment model. Costs and service levels were calculated for each alternative. Three approaches are defined for the ideal replenishment decision. The first approach is the minimum cost strategy; the second is maximizing the service level strategy and the third approach chooses a replenishment strategy that optimizes both the cost and the service level. The replenishment decisions are tested by monitoring the evolution of costs and the service level for the following period. The proposed decision support system can be easily used in ATM management as well as all point of service locations such as vending machines and gas stations.