DEMAND FORECASTING BASED ON MACHINE LEARNING TO DETERMINE ORDER QUANTITY: A CASE STUDY OF BAHAGIA KOPI BANDUNG

Abstract

Coffee shops are becoming increasingly popular in Indonesia, and they are regarded as one of the business sectors that contribute to the country's industrial development. Difficulty to estimate sales and demand, disrupting coffee bean inventory management. Forecasting with machine learning models could provide a solution to these issues. The data used in this study is coffee bean demand from a POS (Point-of-Sales) system, which is calculated by converting coffee menu sales data to coffee bean demand. The data is time-series, spanning from. To improve model effectiveness, several external variables such as weather and event are included. The exploratory data analysis of these factors reveals the influence and pattern that affects the dynamics of coffee bean demand. Prediction models employed in this study include Multiple Linear Regression (MLR), Decision Tree (DT), Support Vector Regressor (SVR), and Neural Network (NN). Model training results demonstrate that models with all variables outperform models with simply date variables. The DT model produces the best forecast based on its pattern and error measurement. The prediction result is executed by constructing a dashboard that assists the businessman in determining the amount of coffee beans to order in the next months. These are the implementations that could be used to improve inventory management.