Perbandingan Algoritma XGBoost dan LSTM untuk Prediksi Ketersediaan Kas Atm Bank Sumut Berbasis Website Geospasial

Abstract

This study compares the performance of the XGBoost and Long Short-Term Memory (LSTM) algorithms in predicting ATM cash availability at Bank Sumut. The predictions are based on historical transaction data processed using machine learning and deep learning methods. XGBoost excels in handling tabular data with interrelated features, while LSTM demonstrates reliability in time series data. The dataset comprises three months of transactions from 20 ATMs located in different areas. The findings indicate that XGBoost achieves higher prediction accuracy than LSTM based on metrics such as RMSE, MAE, and MSE. Furthermore, the prediction implementation is integrated into a geospatial-based website using technologies like Leaflet.js, enabling real-time visualization of ATM cash availability statuses. This system is designed to support Bank Sumut's operational decision-making in efficiently managing cash distribution. The results of this study demonstrate that the XGBoost algorithm outperforms in predicting ATM cash availability, achieving smaller RMSE, MAPE, MAE, and MSE values