Food Product Sales Forecasting Using the LSTM Model: A Case Study of SMAN 1 Malang Teachers’ Cooperative

Abstract

The SMAN 1 Malang Teachers' Cooperative manages consigned food products from local partners but faces fluctuating daily sales and the lack of an accurate forecasting system, which often results in stock shortages or surpluses. This study aims to develop a sales prediction model using a deep learning-based Long Short-Term Memory (LSTM) algorithm to improve forecasting accuracy and procurement efficiency. The developed system utilizes historical transaction data from the cooperative from January 2024 to December 2024, which is converted into time series data. The research stages include data preprocessing with interpolation, rolling mean smoothing, time feature engineering, normalization using MinMaxScaler, and model training using a two-layer LSTM architecture with a fully connected layer (hybrid). Model optimization uses EarlyStopping and ReduceLROnPlateau to prevent overfitting. Model evaluation is conducted separately using the Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE) metrics. The test results show that the optimized hybrid LSTM model achieved an average accuracy of 88.19% with an MAPE value of 11.81%, an MAE of 1.29, an MSE of 6,63, and an RMSE of 1.70. These values indicate that the model has stable, accurate, and adaptive predictive capabilities in estimating daily sales for various consignment food products managed by the SMAN 1 Malang Teachers' Cooperative.