Predictive Analytics in ATM Cash Demand Forecasting Using Machine Learning Approach: A Case Study of the Commercial Bank of Ethiopia
No Thumbnail Available
Date
2026-11-03
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
. Reliable cash availability at Automated Teller Machine (ATM) is important for customer service and for controlling the cost of holding and moving cash. The purpose of this study was to examine if predictive analytics and machine learning models can forecast daily ATM cash demand at CBE and support better replenishment planning. We used quantitative predictive approach to achieve the objective of the study. Secondary data was collected from the bank. Daily ATM transaction records from 15 atms covering two years period were used. Based on the raw data from the bank, data were cleaned preprocessed and new features were created. Three machine learning models were trained and compared namely, Linear Regression as baseline, Random Forest and Extreme Gradient Boosting (xgboost). Model performance was measured using MAE, RMSE, R2 and MAPE with using MAE and R2 as primary metrics. Xgboost was the best model with MAE of 268,821 and R2 of o.6682 and Random Forest was a close second. The analysis showed that recent demand history is the strongest driver of future demand, while salary days, pre-holiday days holiday events also shaped withdrawal. The study concludes that CBE already holds that data needed to move from experience based to data driven cash management and recommends adopting the model as a decision support tool supported by stronger data governance and staff capacity
Description
Keywords
: Predictive analytics, ATM cash demand forecasting