Machine Learning Model using Times Series Analytics for Prediction of ATM Transactions

Automated Teller Machines (ATMs) are banking outlets that support customer activities to process regular financial transactions quickly and automatically without the help of tellers. Transaction activities carried out by customers at ATMs are often difficult to predict. Therefore, the banking sector is strongly encouraged to build an intelligent cash management system to provide opportunities for banks to lower operating costs. Customer transaction patterns are needed to make a prediction. This study aims to implement machine learning algorithms to predict transaction values at ATMs with time series algorithms. This study aims to create a machine learning model to determine the predicted value of ATM transactions using four algorithms, Linear Regression, Prophet, ARIMA, and LSTM algorithms. The dataset used is the data set of ATM transactions of XYZ bank. This dataset has around 11.588 rows and ten columns. The results of the calculations with the best evaluation model determined in the LSTM algorithm, which produces the Mean Absolute Error (MAE) value of 20,686.91, the Mean Squared Error (MSE) of 710,590,544.24, and the Coefficient of Determination (R2) 0.72.