Vending Machine Product Demand Prediction Using Machine Learning Algorithms

The study aims to leverage the vending machine's historic product sales data to predict future demand using machine learning (ML) techniques with the goal of strategic inventory planning, reducing stock waste, and increasing operational efficiency and revenue. The study deploys different ML algorithms including XGBoost, FB Prophet, Autoregressive Integrated Moving Average and Support Vector Regression. This is among the first studies of its kind to explore ML for predictive restocking in the vending industry. The experiment is divided into two parts. In the first part, we used common historic sales data variables to make the prediction where XGBoost emerged as the best-performing model in terms of the mean absolute error. In the second part, we used additional features such as public holidays, day of the week, and sales deviation flags to make the prediction. Using additional variables slightly improved the performance of the ML algorithms with XGBoost having the lowest mean absolute error of 8. Overall, the study demonstrates the efficacy of utilising ML techniques and additional features in accurately predicting future product demand for vending machines, thus enabling strategic inventory planning, minimised stock waste, improved operational efficiency and increased revenue for the vending industry.