Synthetic Data Generation Using Time-Generative Adversarial Network (Time-GAN) to Predict Cash ATM

Abstract

One of the particular concerns of the Bank for ATM transaction services is the availability of cash at ATMs. Prediction of the availability of ATM money is required by the Bank to manage funds optimally. This study aims to analyze and predict cash availability at ATM machines using Time-GAN and Extreme Gradient Boosting (XGBoost). Time-Series data is highly dependent on the size and consistency of the dataset used in the training. The features available in the dataset are limited and have constraints such as missing dimensions or missing values. Therefore, synthetic data generation technique is used as an effective way to increase the amount of data and handle imbalanced data. Synthetic data generation has been shown to increase the generalizability of models with Time-Series data. The generated data will be divided into Training data, Validation data, and Testing data, resulting in a Load Model that will be analyzed using the XGBoost method. The ultimate goal of this research is to provide a summary of the evaluation and performance that results in better ATM availability for future research. Model performance is evaluated with the Mean Absolute Error (MAE) metric 2.57 value, Mean Squared Error (MSE) 1.64 value, and R-squared 5.02 value.