Automated Teller Machines Availability Forecasting Model Using Machine Learning Algorithms رسالة ماجستير

Abstract

Automated Teller Machine is the most widespread bank service, it is also the most widely used among electronic banking services, and it represents an important tool for the bank to spread, promote, and attract clients. There are several factors to take into consideration in ATM cash management, forecasting the withdrawal amount is the first in priority for any bank, replenish with the optimal amount saves expenses and maximizes the profit. Based on raw ATM transactions data from the Bank of Palestine for 2017-2018, these data were preprocessed, and new features were engineered using different approaches, 3 models were utilized (Random Forest, XGBoost, and LSTM), and LSTM gained the best result in tow tasks, first to predict the next-hour withdrawals for 172 ATMs grouped into hourly time series and considered as one ATM, with R2 of 91% and MAPE of 0.37; the other task was to predict the rolling mean transactions amount of the next even hours (2-20), with the best performance of with R2 96%-99% and MAPE of (0.28-0.08). VI This project find out that the cash withdrawal amounts are highly affected by the days of salary transfer of the Palestine National Authority and the hours of the transactions. And demonstrated that utilizing the designed features and suitable machine-learning algorithms allows for accurate predictions of withdrawals from these machines.